feat(web): add Glances integration for host resource stats (#124)

Makes the entry searchable by typing "github" in the command palette.

.envrc.example

@@ -0,0 +1,6 @@
+# Run containers as current user (preserves file ownership on NFS mounts)
+# Copy this file to .envrc and run: direnv allow
+export CF_UID=$(id -u)
+export CF_GID=$(id -g)
+export CF_HOME=$HOME
+export CF_USER=$USER

.gitignore

@@ -37,6 +37,7 @@ ENV/
 .coverage
 .pytest_cache/
 htmlcov/
+.code/
 
 # Local config (don't commit real configs)
 compose-farm.yaml
@@ -45,3 +46,4 @@ coverage.xml
 .env
 homepage/
 site/
+.playwright-mcp/

.pre-commit-config.yaml

@@ -21,7 +21,7 @@ repos:
   - repo: https://github.com/astral-sh/ruff-pre-commit
     rev: v0.14.9
     hooks:
-      - id: ruff
+      - id: ruff-check
         args: [--fix]
       - id: ruff-format
 

.prompts/docs-review.md

@@ -1,118 +1,117 @@
-Review all documentation in this repository for accuracy, completeness, and consistency. Cross-reference documentation against the actual codebase to identify issues.
+Review documentation for accuracy, completeness, and consistency. Focus on things that require judgment—automated checks handle the rest.
 
-## Scope
+## What's Already Automated
 
-Review all documentation files:
-- docs/*.md (primary documentation)
-- README.md (repository landing page)
-- CLAUDE.md (development guidelines)
-- examples/README.md (example configurations)
+Don't waste time on these—CI and pre-commit hooks handle them:
 
-## Review Checklist
+- **README help output**: `markdown-code-runner` regenerates `cf --help` blocks in CI
+- **README command table**: Pre-commit hook verifies commands are listed
+- **Linting/formatting**: Handled by pre-commit
 
-### 1. Command Documentation
+## What This Review Is For
 
-For each documented command, verify against the CLI source code:
+Focus on things that require judgment:
 
-- Command exists in codebase
-- All options are documented with correct names, types, and defaults
-- Short options (-x) match long options (--xxx)
-- Examples would work as written
-- Check for undocumented commands or options
+1. **Accuracy**: Does the documentation match what the code actually does?
+2. **Completeness**: Are there undocumented features, options, or behaviors?
+3. **Clarity**: Would a new user understand this? Are examples realistic?
+4. **Consistency**: Do different docs contradict each other?
+5. **Freshness**: Has the code changed in ways the docs don't reflect?
 
-Run `--help` for each command to verify.
+## Review Process
 
-### 2. Configuration Documentation
+### 1. Check Recent Changes
 
-Verify against Pydantic models in the config module:
+```bash
+# What changed recently that might need doc updates?
+git log --oneline -20 | grep -iE "feat|fix|add|remove|change|option"
 
-- All config keys are documented
-- Types match Pydantic field types
-- Required vs optional fields are correct
-- Default values are accurate
-- Config file search order matches code
-- Example YAML is valid and uses current schema
+# What code files changed?
+git diff --name-only HEAD~20 | grep "\.py$"
+```
 
-### 3. Architecture Documentation
+Look for new features, changed defaults, renamed options, or removed functionality.
 
-Verify against actual directory structure:
+### 2. Verify docs/commands.md Options Tables
 
-- File paths match actual source code location
-- All modules listed actually exist
-- No modules are missing from the list
-- Component descriptions match code functionality
-- CLI module list includes all command files
+The README auto-updates help output, but `docs/commands.md` has **manually maintained options tables**. These can drift.
 
-### 4. State and Data Files
+For each command's options table, compare against `cf <command> --help`:
+- Are all options listed?
+- Are short flags correct?
+- Are defaults accurate?
+- Are descriptions accurate?
 
-Verify against state and path modules:
+**Pay special attention to subcommands** (`cf config *`, `cf ssh *`)—these have their own options that are easy to miss.
 
-- State file name and location are correct
-- State file format matches actual structure
-- Log file name and location are correct
-- What triggers state/log updates is accurate
+### 3. Verify docs/configuration.md
 
-### 5. Installation Documentation
+Compare against Pydantic models in the source:
 
-Verify against pyproject.toml:
+```bash
+# Find the config models
+grep -r "class.*BaseModel" src/ --include="*.py" -A 15
+```
 
-- Python version requirement matches requires-python
-- Package name is correct
-- Optional dependencies are documented
-- CLI entry points are mentioned
-- Installation methods work as documented
+Check:
+- All config keys documented
+- Types and defaults match code
+- Config file search order is accurate
+- Example YAML would actually work
 
-### 6. Feature Claims
+### 4. Verify docs/architecture.md
 
-For each claimed feature, verify it exists and works as described.
+```bash
+# What source files actually exist?
+git ls-files "src/**/*.py"
+```
 
-### 7. Cross-Reference Consistency
+Check:
+- Listed files exist
+- No files are missing from the list
+- Descriptions match what the code does
 
-Check for conflicts between documentation files:
+### 5. Check Examples
 
-- README vs docs/index.md (should be consistent)
-- CLAUDE.md vs actual code structure
-- Command tables match across files
-- Config examples are consistent
+For examples in any doc:
+- Would the YAML/commands actually work?
+- Are service names, paths, and options realistic?
+- Do examples use current syntax (not deprecated options)?
 
-### 8. Recent Changes Check
+### 6. Cross-Reference Consistency
 
-Before starting the review:
+The same info appears in multiple places. Check for conflicts:
+- README.md vs docs/index.md
+- docs/commands.md vs CLAUDE.md command tables
+- Config examples across different docs
 
-- Run `git log --oneline -20` to see recent commits
-- Look for commits with `feat:`, `fix:`, or that mention new options/commands
-- Cross-reference these against the documentation to catch undocumented features
+### 7. Self-Check This Prompt
 
-### 9. Auto-Generated Content
+This prompt can become outdated too. If you notice:
+- New automated checks that should be listed above
+- New doc files that need review guidelines
+- Patterns that caused issues
 
-For README.md or docs with `<!-- CODE:BASH:START -->` blocks:
-
-- Run `uv run markdown-code-runner <file>` to regenerate outputs
-- Check for missing `<!-- OUTPUT:START -->` markers (blocks that never ran)
-- Verify help output matches current CLI behavior
-
-### 10. CLI Options Completeness
-
-For each command, run `cf <command> --help` and verify:
-
-- Every option shown in help is documented
-- Short flags (-x) are listed alongside long flags (--xxx)
-- Default values in help match documented defaults
+Include prompt updates in your fixes.
 
 ## Output Format
 
-Provide findings in these categories:
+Categorize findings:
 
-1. **Critical Issues**: Incorrect information that would cause user problems
-2. **Inaccuracies**: Technical errors, wrong defaults, incorrect paths
-3. **Missing Documentation**: Features/commands that exist but aren't documented
-4. **Outdated Content**: Information that was once true but no longer is
-5. **Inconsistencies**: Conflicts between different documentation files
-6. **Minor Issues**: Typos, formatting, unclear wording
-7. **Verified Accurate**: Sections confirmed to be correct
+1. **Critical**: Wrong info that would break user workflows
+2. **Inaccuracy**: Technical errors (wrong defaults, paths, types)
+3. **Missing**: Undocumented features or options
+4. **Outdated**: Was true, no longer is
+5. **Inconsistency**: Docs contradict each other
+6. **Minor**: Typos, unclear wording
 
-For each issue, include:
-- File path and line number (if applicable)
-- What the documentation says
-- What the code actually does
-- Suggested fix
+For each issue, provide a ready-to-apply fix:
+
+```
+### Issue: [Brief description]
+
+- **File**: docs/commands.md:652
+- **Problem**: `cf ssh setup` has `--config` option but it's not documented
+- **Fix**: Add `--config, -c PATH` to the options table
+- **Verify**: `cf ssh setup --help`
+```

.prompts/pr-review.md

@@ -0,0 +1,15 @@
+Review the pull request for:
+
+- **Code cleanliness**: Is the implementation clean and well-structured?
+- **DRY principle**: Does it avoid duplication?
+- **Code reuse**: Are there parts that should be reused from other places?
+- **Organization**: Is everything in the right place?
+- **Consistency**: Is it in the same style as other parts of the codebase?
+- **Simplicity**: Is it not over-engineered? Remember KISS and YAGNI. No dead code paths and NO defensive programming.
+- **User experience**: Does it provide a good user experience?
+- **PR**: Is the PR description and title clear and informative?
+- **Tests**: Are there tests, and do they cover the changes adequately? Are they testing something meaningful or are they just trivial?
+- **Live tests**: Test the changes in a REAL live environment to ensure they work as expected, use the config in `/opt/stacks/compose-farm.yaml`.
+- **Rules**: Does the code follow the project's coding standards and guidelines as laid out in @CLAUDE.md?
+
+Look at `git diff origin/main..HEAD` for the changes made in this pull request.

.prompts/update-demos.md

@@ -0,0 +1,51 @@
+Update demo recordings to match the current compose-farm.yaml configuration.
+
+## Key Gotchas
+
+1. **Never `git checkout` without asking** - check for uncommitted changes first
+2. **Prefer `nas` stacks** - demos run locally on nas, SSH adds latency
+3. **Terminal captures keyboard** - use `blur()` to release focus before command palette
+4. **Clicking sidebar navigates away** - clicking h1 scrolls to top
+5. **Buttons have icons, not text** - use `[data-tip="..."]` selectors
+6. **`record.py` auto-restores config** - no manual cleanup needed after CLI demos
+
+## Stacks Used in Demos
+
+| Stack | CLI Demos | Web Demos | Notes |
+|-------|-----------|-----------|-------|
+| `audiobookshelf` | quickstart, migration, apply | - | Migrates nas→anton |
+| `grocy` | update | navigation, stack, workflow, console | - |
+| `immich` | logs, compose | shell | Multiple containers |
+| `dozzle` | - | workflow | - |
+
+## CLI Demos
+
+**Files:** `docs/demos/cli/*.tape`
+
+Check:
+- `quickstart.tape`: `bat -r` line ranges match current config structure
+- `migration.tape`: nvim keystrokes work, stack exists on nas
+- `compose.tape`: exec commands produce meaningful output
+
+Run: `python docs/demos/cli/record.py [demo]`
+
+## Web Demos
+
+**Files:** `docs/demos/web/demo_*.py`
+
+Check:
+- Stack names in demos still exist in config
+- Selectors match current templates (grep for IDs in `templates/`)
+- Shell demo uses command palette for ALL navigation
+
+Run: `python docs/demos/web/record.py [demo]`
+
+## Before Recording
+
+```bash
+# Check for uncommitted config changes
+git -C /opt/stacks diff compose-farm.yaml
+
+# Verify stacks are running
+cf ps audiobookshelf grocy immich dozzle
+```

Dockerfile

@@ -16,5 +16,13 @@ RUN apk add --no-cache openssh-client
 COPY --from=builder /root/.local/share/uv/tools/compose-farm /root/.local/share/uv/tools/compose-farm
 COPY --from=builder /usr/local/bin/cf /usr/local/bin/compose-farm /usr/local/bin/
 
-ENTRYPOINT ["cf"]
+# Allow non-root users to access the installed tool
+# (required when running with user: "${CF_UID:-0}:${CF_GID:-0}")
+RUN chmod 755 /root
+
+# Allow non-root users to add passwd entries (required for SSH)
+RUN chmod 666 /etc/passwd
+
+# Entrypoint creates /etc/passwd entry for non-root UIDs (required for SSH)
+ENTRYPOINT ["sh", "-c", "[ $(id -u) != 0 ] && echo ${USER:-u}:x:$(id -u):$(id -g)::${HOME:-/}:/bin/sh >> /etc/passwd; exec cf \"$@\"", "--"]
 CMD ["--help"]

README.md

@@ -5,12 +5,19 @@
 [![License](https://img.shields.io/github/license/basnijholt/compose-farm)](LICENSE)
 [![GitHub stars](https://img.shields.io/github/stars/basnijholt/compose-farm)](https://github.com/basnijholt/compose-farm/stargazers)
 
-<img src="http://files.nijho.lt/compose-farm.png" align="right" style="width: 300px;" />
+<img src="https://files.nijho.lt/compose-farm.png" alt="Compose Farm logo" align="right" style="width: 300px;" />
 
 A minimal CLI tool to run Docker Compose commands across multiple hosts via SSH.
 
 > [!NOTE]
-> Run `docker compose` commands across multiple hosts via SSH. One YAML maps stacks to hosts. Run `cf apply` and reality matches your config—stacks start, migrate, or stop as needed. No Kubernetes, no Swarm, no magic.
+> Agentless multi-host Docker Compose. CLI-first with a web UI. Your files stay as plain folders—version-controllable, no lock-in. Run `cf apply` and reality matches your config.
+
+**Why Compose Farm?**
+- **Your files, your control** — Plain folders + YAML, not locked in Portainer. Version control everything.
+- **Agentless** — Just SSH, no agents to deploy (unlike [Dockge](https://github.com/louislam/dockge)).
+- **Zero changes required** — Existing compose files work as-is.
+- **Grows with you** — Start single-host, scale to multi-host seamlessly.
+- **Declarative** — Change config, run `cf apply`, reality matches.
 
 ## Quick Demo
 
@@ -36,8 +43,8 @@ A minimal CLI tool to run Docker Compose commands across multiple hosts via SSH.
   - [What Compose Farm doesn't do](#what-compose-farm-doesnt-do)
 - [Installation](#installation)
 - [SSH Authentication](#ssh-authentication)
-  - [SSH Agent (default)](#ssh-agent-default)
-  - [Dedicated SSH Key (recommended for Docker/Web UI)](#dedicated-ssh-key-recommended-for-dockerweb-ui)
+  - [SSH Agent](#ssh-agent)
+  - [Dedicated SSH Key (default for Docker)](#dedicated-ssh-key-default-for-docker)
 - [Configuration](#configuration)
   - [Single-host example](#single-host-example)
   - [Multi-host example](#multi-host-example)
@@ -47,6 +54,7 @@ A minimal CLI tool to run Docker Compose commands across multiple hosts via SSH.
   - [CLI `--help` Output](#cli---help-output)
   - [Auto-Migration](#auto-migration)
 - [Traefik Multihost Ingress (File Provider)](#traefik-multihost-ingress-file-provider)
+- [Host Resource Monitoring (Glances)](#host-resource-monitoring-glances)
 - [Comparison with Alternatives](#comparison-with-alternatives)
 - [License](#license)
 
@@ -177,15 +185,33 @@ docker run --rm \
   ghcr.io/basnijholt/compose-farm up --all
 ```
 
+**Running as non-root user** (recommended for NFS mounts):
+
+By default, containers run as root. To preserve file ownership on mounted volumes
+(e.g., `compose-farm-state.yaml`, config edits), set these environment variables:
+
+```bash
+# Add to .env file (one-time setup)
+echo "CF_UID=$(id -u)" >> .env
+echo "CF_GID=$(id -g)" >> .env
+echo "CF_HOME=$HOME" >> .env
+echo "CF_USER=$USER" >> .env
+```
+
+Or use [direnv](https://direnv.net/) (copies `.envrc.example` to `.envrc`):
+```bash
+cp .envrc.example .envrc && direnv allow
+```
+
 </details>
 
 ## SSH Authentication
 
 Compose Farm uses SSH to run commands on remote hosts. There are two authentication methods:
 
-### SSH Agent (default)
+### SSH Agent
 
-Works out of the box if you have an SSH agent running with your keys loaded:
+Works out of the box when running locally if you have an SSH agent running with your keys loaded:
 
 ```bash
 # Verify your agent has keys
@@ -195,9 +221,9 @@ ssh-add -l
 cf up --all
 ```
 
-### Dedicated SSH Key (recommended for Docker/Web UI)
+### Dedicated SSH Key (default for Docker)
 
-When running compose-farm in Docker, the SSH agent connection can be lost (e.g., after container restart). The `cf ssh` command sets up a dedicated key that persists:
+When running in Docker, SSH agent sockets are ephemeral and can be lost after container restarts. The `cf ssh` command sets up a dedicated key that persists:
 
 ```bash
 # Generate key and copy to all configured hosts
@@ -216,15 +242,22 @@ When running in Docker, mount a volume to persist the SSH keys. Choose ONE optio
 **Option 1: Host path (default)** - keys at `~/.ssh/compose-farm/id_ed25519`
 ```yaml
 volumes:
-  - ~/.ssh/compose-farm:/root/.ssh
+  - ~/.ssh/compose-farm:${CF_HOME:-/root}/.ssh
 ```
 
 **Option 2: Named volume** - managed by Docker
 ```yaml
 volumes:
-  - cf-ssh:/root/.ssh
+  - cf-ssh:${CF_HOME:-/root}/.ssh
 ```
 
+**Option 3: SSH agent forwarding** - if you prefer using your host's ssh-agent
+```yaml
+volumes:
+  - ${SSH_AUTH_SOCK}:/ssh-agent:ro
+```
+Note: Requires `SSH_AUTH_SOCK` environment variable to be set. The socket path is ephemeral and changes across sessions.
+
 Run setup once after starting the container (while the SSH agent still works):
 
 ```bash
@@ -233,6 +266,8 @@ docker compose exec web cf ssh setup
 
 The keys will persist across restarts.
 
+**Note:** When running as non-root (with `CF_UID`/`CF_GID`), set `CF_HOME` to your home directory so SSH finds the keys at the correct path.
+
 </details>
 
 ## Configuration
@@ -422,6 +457,15 @@ Full `--help` output for each command. See the [Usage](#usage) table above for a
 │                                 copy it or customize the installation.       │
 │ --help                -h        Show this message and exit.                  │
 ╰──────────────────────────────────────────────────────────────────────────────╯
+╭─ Configuration ──────────────────────────────────────────────────────────────╮
+│ traefik-file   Generate a Traefik file-provider fragment from compose        │
+│                Traefik labels.                                               │
+│ refresh        Update local state from running stacks.                       │
+│ check          Validate configuration, traefik labels, mounts, and networks. │
+│ init-network   Create Docker network on hosts with consistent settings.      │
+│ config         Manage compose-farm configuration files.                      │
+│ ssh            Manage SSH keys for passwordless authentication.              │
+╰──────────────────────────────────────────────────────────────────────────────╯
 ╭─ Lifecycle ──────────────────────────────────────────────────────────────────╮
 │ up             Start stacks (docker compose up -d). Auto-migrates if host    │
 │                changed.                                                      │
@@ -433,18 +477,10 @@ Full `--help` output for each command. See the [Usage](#usage) table above for a
 │                that service.                                                 │
 │ update         Update stacks (pull + build + down + up). With --service,     │
 │                updates just that service.                                    │
-│ apply          Make reality match config (start, migrate, stop as needed).   │
+│ apply          Make reality match config (start, migrate, stop               │
+│                strays/orphans as needed).                                    │
 │ compose        Run any docker compose command on a stack.                    │
 ╰──────────────────────────────────────────────────────────────────────────────╯
-╭─ Configuration ──────────────────────────────────────────────────────────────╮
-│ traefik-file   Generate a Traefik file-provider fragment from compose        │
-│                Traefik labels.                                               │
-│ refresh        Update local state from running stacks.                       │
-│ check          Validate configuration, traefik labels, mounts, and networks. │
-│ init-network   Create Docker network on hosts with consistent settings.      │
-│ config         Manage compose-farm configuration files.                      │
-│ ssh            Manage SSH keys for passwordless authentication.              │
-╰──────────────────────────────────────────────────────────────────────────────╯
 ╭─ Monitoring ─────────────────────────────────────────────────────────────────╮
 │ logs           Show stack logs. With --service, shows logs for just that     │
 │                service.                                                      │
@@ -694,22 +730,25 @@ Full `--help` output for each command. See the [Usage](#usage) table above for a
 
  Usage: cf apply [OPTIONS]
 
- Make reality match config (start, migrate, stop as needed).
+ Make reality match config (start, migrate, stop strays/orphans as needed).
 
  This is the "reconcile" command that ensures running stacks match your
  config file. It will:
 
  1. Stop orphaned stacks (in state but removed from config)
- 2. Migrate stacks on wrong host (host in state ≠ host in config)
- 3. Start missing stacks (in config but not in state)
+ 2. Stop stray stacks (running on unauthorized hosts)
+ 3. Migrate stacks on wrong host (host in state ≠ host in config)
+ 4. Start missing stacks (in config but not in state)
 
  Use --dry-run to preview changes before applying.
- Use --no-orphans to only migrate/start without stopping orphaned stacks.
+ Use --no-orphans to skip stopping orphaned stacks.
+ Use --no-strays to skip stopping stray stacks.
  Use --full to also run 'up' on all stacks (picks up compose/env changes).
 
 ╭─ Options ────────────────────────────────────────────────────────────────────╮
 │ --dry-run     -n            Show what would change without executing         │
 │ --no-orphans                Only migrate, don't stop orphaned stacks         │
+│ --no-strays                 Don't stop stray stacks (running on wrong host)  │
 │ --full        -f            Also run up on all stacks to apply config        │
 │                             changes                                          │
 │ --config      -c      PATH  Path to config file                              │
@@ -1287,6 +1326,52 @@ Update your Traefik config to use directory watching instead of a single file:
 - --providers.file.watch=true
 ```
 
+## Host Resource Monitoring (Glances)
+
+The web UI can display real-time CPU, memory, and load stats for all configured hosts. This uses [Glances](https://nicolargo.github.io/glances/), a cross-platform system monitoring tool with a REST API.
+
+**Setup**
+
+1. Deploy a Glances stack that runs on all hosts:
+
+```yaml
+# glances/compose.yaml
+name: glances
+services:
+  glances:
+    image: nicolargo/glances:latest
+    container_name: glances
+    restart: unless-stopped
+    pid: host
+    ports:
+      - "61208:61208"
+    environment:
+      - GLANCES_OPT=-w  # Enable web server mode
+    volumes:
+      - /var/run/docker.sock:/var/run/docker.sock:ro
+```
+
+2. Add it to your config as a multi-host stack:
+
+```yaml
+# compose-farm.yaml
+stacks:
+  glances: all  # Runs on every host
+
+glances_stack: glances  # Enables resource stats in web UI
+```
+
+3. Deploy: `cf up glances`
+
+The web UI dashboard will now show a "Host Resources" section with live stats from all hosts. Hosts where Glances is unreachable show an error indicator.
+
+**Live Stats Page**
+
+With Glances configured, a Live Stats page (`/live-stats`) shows all running containers across all hosts:
+
+- **Columns**: Stack, Service, Host, Image, Status, Uptime, CPU, Memory, Net I/O
+- **Features**: Sorting, filtering, live updates (no SSH required—uses Glances REST API)
+
 ## Comparison with Alternatives
 
 There are many ways to run containers on multiple hosts. Here is where Compose Farm sits:

docker-compose.yml

@@ -1,38 +1,56 @@
 services:
   cf:
     image: ghcr.io/basnijholt/compose-farm:latest
+    # Run as current user to preserve file ownership on mounted volumes
+    # Set CF_UID=$(id -u) CF_GID=$(id -g) in your environment or .env file
+    # Defaults to root (0:0) for backwards compatibility
+    user: "${CF_UID:-0}:${CF_GID:-0}"
     volumes:
-      - ${SSH_AUTH_SOCK}:/ssh-agent:ro
       # Compose directory (contains compose files AND compose-farm.yaml config)
       - ${CF_COMPOSE_DIR:-/opt/stacks}:${CF_COMPOSE_DIR:-/opt/stacks}
       # SSH keys for passwordless auth (generated by `cf ssh setup`)
       # Choose ONE option below (use the same option for both cf and web services):
       # Option 1: Host path (default) - keys at ~/.ssh/compose-farm/id_ed25519
-      - ${CF_SSH_DIR:-~/.ssh/compose-farm}:/root/.ssh
+      - ${CF_SSH_DIR:-~/.ssh/compose-farm}:${CF_HOME:-/root}/.ssh/compose-farm
       # Option 2: Named volume - managed by Docker, shared between services
-      # - cf-ssh:/root/.ssh
+      # - cf-ssh:${CF_HOME:-/root}/.ssh
+      # Option 3: SSH agent forwarding (uncomment if using ssh-agent)
+      # - ${SSH_AUTH_SOCK}:/ssh-agent:ro
     environment:
       - SSH_AUTH_SOCK=/ssh-agent
       # Config file path (state stored alongside it)
       - CF_CONFIG=${CF_COMPOSE_DIR:-/opt/stacks}/compose-farm.yaml
+      # HOME must match the user running the container for SSH to find keys
+      - HOME=${CF_HOME:-/root}
+      # USER is required for SSH when running as non-root (UID not in /etc/passwd)
+      - USER=${CF_USER:-root}
 
   web:
     image: ghcr.io/basnijholt/compose-farm:latest
     restart: unless-stopped
     command: web --host 0.0.0.0 --port 9000
+    # Run as current user to preserve file ownership on mounted volumes
+    user: "${CF_UID:-0}:${CF_GID:-0}"
     volumes:
-      - ${SSH_AUTH_SOCK}:/ssh-agent:ro
       - ${CF_COMPOSE_DIR:-/opt/stacks}:${CF_COMPOSE_DIR:-/opt/stacks}
       # SSH keys - use the SAME option as cf service above
       # Option 1: Host path (default)
-      - ${CF_SSH_DIR:-~/.ssh/compose-farm}:/root/.ssh
+      - ${CF_SSH_DIR:-~/.ssh/compose-farm}:${CF_HOME:-/root}/.ssh/compose-farm
       # Option 2: Named volume
-      # - cf-ssh:/root/.ssh
+      # - cf-ssh:${CF_HOME:-/root}/.ssh
+      # Option 3: SSH agent forwarding (uncomment if using ssh-agent)
+      # - ${SSH_AUTH_SOCK}:/ssh-agent:ro
+      # XDG config dir for backups and image digest logs (persists across restarts)
+      - ${CF_XDG_CONFIG:-~/.config/compose-farm}:${CF_HOME:-/root}/.config/compose-farm
     environment:
       - SSH_AUTH_SOCK=/ssh-agent
       - CF_CONFIG=${CF_COMPOSE_DIR:-/opt/stacks}/compose-farm.yaml
       # Used to detect self-updates and run via SSH to survive container restart
       - CF_WEB_STACK=compose-farm
+      # HOME must match the user running the container for SSH to find keys
+      - HOME=${CF_HOME:-/root}
+      # USER is required for SSH when running as non-root (UID not in /etc/passwd)
+      - USER=${CF_USER:-root}
     labels:
       - traefik.enable=true
       - traefik.http.routers.compose-farm.rule=Host(`compose-farm.${DOMAIN}`)

docs/assets/apply.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb1372a59a4ed1ac74d3864d7a84dd5311fce4cb6c6a00bf3a574bc2f98d5595
-size 895927
+oid sha256:01dabdd8f62773823ba2b8dc74f9931f1a1b88215117e6a080004096025491b0
+size 901456

docs/assets/apply.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f339a85f3d930db5a020c9f77e106edc5f44ea7dee6f68557106721493c24ef8
-size 205907
+oid sha256:134c903a6b3acfb933617b33755b0cdb9bac2a59e5e35b64236e248a141d396d
+size 206883

docs/assets/compose.gif

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d8b3cdb3486ec79b3ddb2f7571c13d54ac9aed182edfe708eff76a966a90cfc7
+size 1132310

docs/assets/compose.webm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a3c4d4a62f062f717df4e6752efced3caea29004dc90fe97fd7633e7f0ded9db
+size 341057

docs/assets/install.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:388aa49a1269145698f9763452aaf6b9c6232ea9229abe1dae304df558e29695
-size 403442
+oid sha256:6c1bb48cc2f364681515a4d8bd0c586d133f5a32789b7bb64524ad7d9ed0a8e9
+size 543135

docs/assets/install.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9b8bf4dcb8ee67270d4a88124b4dd4abe0dab518e73812ee73f7c66d77f146e2
-size 228025
+oid sha256:5f82d96137f039f21964c15c1550aa1b1f0bb2d52c04d012d253dbfbd6fad096
+size 268086

docs/assets/logs.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:16b9a28137dfae25488e2094de85766a039457f5dca20c2d84ac72e3967c10b9
-size 164237
+oid sha256:2a4045b00d90928f42c7764b3c24751576cfb68a34c6e84d12b4e282d2e67378
+size 146467

docs/assets/logs.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0fbe697a1f8256ce3b9a6a64c7019d42769134df9b5b964e5abe98a29e918fd
-size 68242
+oid sha256:f1b94416ed3740853f863e19bf45f26241a203fb0d7d187160a537f79aa544fa
+size 60353

docs/assets/migration.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:629b8c80b98eb996b75439745676fd99a83f391ca25f778a71bd59173f814c2f
-size 1194931
+oid sha256:848d9c48fb7511da7996149277c038589fad1ee406ff2f30c28f777fc441d919
+size 1183641

docs/assets/migration.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33fd46f2d8538cc43be4cb553b3af9d8b412f282ee354b6373e2793fe41c799b
-size 405057
+oid sha256:e747ee71bb38b19946005d5a4def4d423dadeaaade452dec875c4cb2d24a5b77
+size 407373

docs/assets/quickstart.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ccd96e33faba5f297999917d89834b29d58bd2a8929eea8d62875e3d8830bd5c
-size 3198466
+oid sha256:d32c9a3eec06e57df085ad347e6bf61e323f8bd8322d0c540f0b9d4834196dfd
+size 3589776

docs/assets/quickstart.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:979a1a21303bbf284b3510981066ef05c41c1035b34392fecc7bee472116e6db
-size 967564
+oid sha256:6c54eda599389dac74c24c83527f95cd1399e653d7faf2972c2693d90e590597
+size 1085344

docs/assets/update.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2067f4967a93b7ee3a8db7750c435f41b1fccd2919f3443da4b848c20cc54f23
-size 124559
+oid sha256:62f9b5ec71496197a3f1c3e3bca8967d603838804279ea7dbf00a70d3391ff6c
+size 127123

docs/assets/update.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5471bd94e6d1b9d415547fa44de6021fdad2e1cc5b8b295680e217104aa749d6
-size 98149
+oid sha256:ac2b93d3630af87b44a135723c5d10e8287529bed17c28301b2802cd9593e9e8
+size 98748

docs/assets/web-console.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dac5660cfe6574857ec055fac7822f25b7c5fcb10a836b19c86142515e2fbf75
-size 1816075
+oid sha256:7b50a7e9836c496c0989363d1440fa0a6ccdaa38ee16aae92b389b3cf3c3732f
+size 2385110

docs/assets/web-console.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d4efec8ef5a99f2cb31d55cd71cdbf0bb8dd0cd6281571886b7c1f8b41c3f9da
-size 1660764
+oid sha256:ccbb3d5366c7734377e12f98cca0b361028f5722124f1bb7efa231f6aeffc116
+size 2208044

docs/assets/web-live_stats.gif

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4135888689a10c5ae2904825d98f2a6d215c174a4bd823e25761f619590f04ff
+size 3990104

docs/assets/web-live_stats.webm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87739cd6f6576a81100392d8d1e59d3e776fecc8f0721a31332df89e7fc8593d
+size 5814274

docs/assets/web-navigation.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9348dd36e79192344476d61fbbffdb122a96ecc5829fbece1818590cfc521521
-size 3373003
+oid sha256:269993b52721ce70674d3aab2a4cd8c58aa621d4ba0739afedae661c90965b26
+size 3678371

docs/assets/web-navigation.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bebbf8151434ba37bf5e46566a4e8b57812944281926f579d056bdc835ca26aa
-size 2729799
+oid sha256:0098b55bb6a52fa39f807a01fa352ce112bcb446e2a2acb963fb02d21b28c934
+size 3088813

docs/assets/web-shell.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3712afff6fcde00eb951264bb24d4301deb085d082b4e95ed4c1893a571938ee
-size 1528294
+oid sha256:4bf9d8c247d278799d1daea784fc662a22f12b1bd7883f808ef30f35025ebca6
+size 4166443

docs/assets/web-shell.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b218d400836a50661c9cdcce2d2b1e285cc5fe592cb42f58aae41f3e7d60684
-size 1327413
+oid sha256:02d5124217a94849bf2971d6d13d28da18c557195a81b9cca121fb7c07f0501b
+size 3523244

docs/assets/web-stack.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6a232ddc1b9ddd9bf6b5d99c05153e1094be56f1952f02636ca498eb7484e096
-size 3808675
+oid sha256:412a0e68f8e52801cafbb9a703ca9577e7c14cc7c0e439160b9185961997f23c
+size 4435697

docs/assets/web-stack.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a7c9f5f6d47074a6af135190fda6d0a1936cd7a0b04b3aa04ea7d99167a9e05
-size 3333014
+oid sha256:0e600a1d3216b44497a889f91eac94d62ef7207b4ed0471465dcb72408caa28e
+size 3764693

docs/assets/web-themes.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:66f4547ed2e83b302d795875588d9a085af76071a480f1096f2bb64344b80c42
-size 5428670
+oid sha256:3c07a283f4f70c4ab205b0f0acb5d6f55e3ced4c12caa7a8d5914ffe3548233a
+size 5768166

docs/assets/web-themes.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:75c8cdeefbbdcab2a240821d3410539f2a2cbe0a015897f4135404c80c3ac32c
-size 6578366
+oid sha256:562228841de976d70ee80999b930eadf3866a13ff2867d900279993744c44671
+size 6667918

docs/assets/web-workflow.gif

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff2e3ca5a46397efcd5f3a595e7d3c179266cc4f3f5f528b428f5ef2a423028e
-size 12649149
+oid sha256:845746ac1cb101c3077d420c4f3fda3ca372492582dc123ac8a031a68ae9b6b1
+size 12943150

docs/assets/web-workflow.webm

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2d739c5f77ddd9d90b609e31df620b35988081b7341fe225eb717d71a87caa88
-size 12284953
+oid sha256:189259558b5760c02583885168d7b0b47cf476cba81c7c028ec770f9d6033129
+size 12415357

docs/commands.md

@@ -290,6 +290,10 @@ cf pull immich --service database
 
 Run any docker compose command on a stack. This is a passthrough to docker compose for commands not wrapped by cf.
 
+<video autoplay loop muted playsinline>
+  <source src="/assets/compose.webm" type="video/webm">
+</video>
+
 ```bash
 cf compose [OPTIONS] STACK COMMAND [ARGS]...
 ```
@@ -649,7 +653,20 @@ cf ssh COMMAND
 | `status` | Show SSH key status and host connectivity |
 | `keygen` | Generate key without distributing |
 
-**Options for `cf ssh setup` and `cf ssh keygen`:**
+**Options for `cf ssh setup`:**
+
+| Option | Description |
+|--------|-------------|
+| `--config, -c PATH` | Path to config file |
+| `--force, -f` | Regenerate key even if it exists |
+
+**Options for `cf ssh status`:**
+
+| Option | Description |
+|--------|-------------|
+| `--config, -c PATH` | Path to config file |
+
+**Options for `cf ssh keygen`:**
 
 | Option | Description |
 |--------|-------------|

docs/demos/cli/README.md

@@ -10,10 +10,10 @@ VHS-based terminal demo recordings for Compose Farm CLI.
 
 ```bash
 # Record all demos
-./docs/demos/cli/record.sh
+python docs/demos/cli/record.py
 
-# Record single demo
-cd /opt/stacks && vhs docs/demos/cli/quickstart.tape
+# Record specific demos
+python docs/demos/cli/record.py quickstart migration
 ```
 
 ## Demos
@@ -23,6 +23,7 @@ cd /opt/stacks && vhs docs/demos/cli/quickstart.tape
 | `install.tape` | Installing with `uv tool install` |
 | `quickstart.tape` | `cf ps`, `cf up`, `cf logs` |
 | `logs.tape` | Viewing logs |
+| `compose.tape` | `cf compose` passthrough (--help, images, exec) |
 | `update.tape` | `cf update` |
 | `migration.tape` | Service migration |
 | `apply.tape` | `cf apply` |

docs/demos/cli/compose.tape

@@ -0,0 +1,50 @@
+# Compose Demo
+# Shows that cf compose passes through ANY docker compose command
+
+Output docs/assets/compose.gif
+Output docs/assets/compose.webm
+
+Set Shell "bash"
+Set FontSize 14
+Set Width 900
+Set Height 550
+Set Theme "Catppuccin Mocha"
+Set TypingSpeed 50ms
+
+Type "# cf compose runs ANY docker compose command on the right host"
+Enter
+Sleep 500ms
+
+Type "# See ALL available compose commands"
+Enter
+Sleep 500ms
+
+Type "cf compose immich --help"
+Enter
+Sleep 4s
+
+Type "# Show images"
+Enter
+Sleep 500ms
+
+Type "cf compose immich images"
+Enter
+Wait+Screen /immich/
+Sleep 2s
+
+Type "# Open shell in a container"
+Enter
+Sleep 500ms
+
+Type "cf compose immich exec immich-machine-learning sh"
+Enter
+Wait+Screen /#/
+Sleep 1s
+
+Type "python3 --version"
+Enter
+Sleep 1s
+
+Type "exit"
+Enter
+Sleep 500ms

docs/demos/cli/quickstart.tape

@@ -21,7 +21,7 @@ Type "# First, define your hosts..."
 Enter
 Sleep 500ms
 
-Type "bat -r 1:11 compose-farm.yaml"
+Type "bat -r 1:16 compose-farm.yaml"
 Enter
 Sleep 3s
 Type "q"
@@ -31,7 +31,7 @@ Type "# Then map each stack to a host"
 Enter
 Sleep 500ms
 
-Type "bat -r 13:30 compose-farm.yaml"
+Type "bat -r 17:35 compose-farm.yaml"
 Enter
 Sleep 3s
 Type "q"

docs/demos/cli/record.py

@@ -0,0 +1,134 @@
+#!/usr/bin/env python3
+"""Record CLI demos using VHS."""
+
+import shutil
+import subprocess
+import sys
+from pathlib import Path
+
+from rich.console import Console
+
+from compose_farm.config import load_config
+from compose_farm.state import load_state
+
+console = Console()
+SCRIPT_DIR = Path(__file__).parent
+STACKS_DIR = Path("/opt/stacks")
+CONFIG_FILE = STACKS_DIR / "compose-farm.yaml"
+OUTPUT_DIR = SCRIPT_DIR.parent.parent / "assets"
+
+DEMOS = ["install", "quickstart", "logs", "compose", "update", "migration", "apply"]
+
+
+def _run(cmd: list[str], **kw) -> bool:
+    return subprocess.run(cmd, check=False, **kw).returncode == 0
+
+
+def _set_config(host: str) -> None:
+    """Set audiobookshelf host in config file."""
+    _run(["sed", "-i", f"s/audiobookshelf: .*/audiobookshelf: {host}/", str(CONFIG_FILE)])
+
+
+def _get_hosts() -> tuple[str | None, str | None]:
+    """Return (config_host, state_host) for audiobookshelf."""
+    config = load_config()
+    state = load_state(config)
+    return config.stacks.get("audiobookshelf"), state.get("audiobookshelf")
+
+
+def _setup_state(demo: str) -> bool:
+    """Set up required state for demo. Returns False on failure."""
+    if demo not in ("migration", "apply"):
+        return True
+
+    config_host, state_host = _get_hosts()
+
+    if demo == "migration":
+        # Migration needs audiobookshelf on nas in BOTH config and state
+        if config_host != "nas":
+            console.print("[yellow]Setting up: config → nas[/yellow]")
+            _set_config("nas")
+        if state_host != "nas":
+            console.print("[yellow]Setting up: state → nas[/yellow]")
+            if not _run(["cf", "apply"], cwd=STACKS_DIR):
+                return False
+
+    elif demo == "apply":
+        # Apply needs config=nas, state=anton (so there's something to apply)
+        if config_host != "nas":
+            console.print("[yellow]Setting up: config → nas[/yellow]")
+            _set_config("nas")
+        if state_host == "nas":
+            console.print("[yellow]Setting up: state → anton[/yellow]")
+            _set_config("anton")
+            if not _run(["cf", "apply"], cwd=STACKS_DIR):
+                return False
+            _set_config("nas")
+
+    return True
+
+
+def _record(name: str, index: int, total: int) -> bool:
+    """Record a single demo."""
+    console.print(f"[cyan][{index}/{total}][/cyan] [green]Recording:[/green] {name}")
+    if _run(["vhs", str(SCRIPT_DIR / f"{name}.tape")], cwd=STACKS_DIR):
+        console.print("[green]  ✓ Done[/green]")
+        return True
+    console.print("[red]  ✗ Failed[/red]")
+    return False
+
+
+def _reset_after(demo: str, next_demo: str | None) -> None:
+    """Reset state after demos that modify audiobookshelf."""
+    if demo not in ("quickstart", "migration"):
+        return
+    _set_config("nas")
+    if next_demo != "apply":  # Let apply demo show the migration
+        _run(["cf", "apply"], cwd=STACKS_DIR)
+
+
+def _restore_config(original: str) -> None:
+    """Restore original config and sync state."""
+    console.print("[yellow]Restoring original config...[/yellow]")
+    CONFIG_FILE.write_text(original)
+    _run(["cf", "apply"], cwd=STACKS_DIR)
+
+
+def _main() -> int:
+    if not shutil.which("vhs"):
+        console.print("[red]VHS not found. Install: brew install vhs[/red]")
+        return 1
+
+    if not _run(["git", "-C", str(STACKS_DIR), "diff", "--quiet", "compose-farm.yaml"]):
+        console.print("[red]compose-farm.yaml has uncommitted changes[/red]")
+        return 1
+
+    demos = [d for d in sys.argv[1:] if d in DEMOS] or DEMOS
+    if sys.argv[1:] and not demos:
+        console.print(f"[red]Unknown demo. Available: {', '.join(DEMOS)}[/red]")
+        return 1
+
+    # Save original config to restore after recording
+    original_config = CONFIG_FILE.read_text()
+
+    try:
+        for i, demo in enumerate(demos, 1):
+            if not _setup_state(demo):
+                return 1
+            if not _record(demo, i, len(demos)):
+                return 1
+            _reset_after(demo, demos[i] if i < len(demos) else None)
+    finally:
+        _restore_config(original_config)
+
+    # Move outputs
+    OUTPUT_DIR.mkdir(exist_ok=True)
+    for f in (STACKS_DIR / "docs/assets").glob("*.[gw]*"):
+        shutil.move(str(f), str(OUTPUT_DIR / f.name))
+
+    console.print(f"\n[green]Done![/green] Saved to {OUTPUT_DIR}")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(_main())

docs/demos/cli/record.sh

@@ -1,89 +0,0 @@
-#!/usr/bin/env bash
-# Record all VHS demos
-# Run this on a Docker host with compose-farm configured
-
-set -euo pipefail
-
-SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
-DEMOS_DIR="$(dirname "$SCRIPT_DIR")"
-DOCS_DIR="$(dirname "$DEMOS_DIR")"
-REPO_DIR="$(dirname "$DOCS_DIR")"
-OUTPUT_DIR="$DOCS_DIR/assets"
-
-# Colors
-GREEN='\033[0;32m'
-BLUE='\033[0;34m'
-YELLOW='\033[0;33m'
-RED='\033[0;31m'
-NC='\033[0m' # No Color
-
-# Check for VHS
-if ! command -v vhs &> /dev/null; then
-    echo "VHS not found. Install with:"
-    echo "  brew install vhs"
-    echo "  # or"
-    echo "  go install github.com/charmbracelet/vhs@latest"
-    exit 1
-fi
-
-# Ensure output directory exists
-mkdir -p "$OUTPUT_DIR"
-
-# Temp output dir (VHS runs from /opt/stacks, so relative paths go here)
-TEMP_OUTPUT="/opt/stacks/docs/assets"
-mkdir -p "$TEMP_OUTPUT"
-
-# Change to /opt/stacks so cf commands use installed version (not editable install)
-cd /opt/stacks
-
-# Ensure compose-farm.yaml has no uncommitted changes (safety check)
-if ! git diff --quiet compose-farm.yaml; then
-    echo -e "${RED}Error: compose-farm.yaml has uncommitted changes${NC}"
-    echo "Commit or stash your changes before recording demos"
-    exit 1
-fi
-
-echo -e "${BLUE}Recording VHS demos...${NC}"
-echo "Output directory: $OUTPUT_DIR"
-echo ""
-
-# Function to record a tape
-record_tape() {
-    local tape=$1
-    local name=$(basename "$tape" .tape)
-    echo -e "${GREEN}Recording:${NC} $name"
-    if vhs "$tape"; then
-        echo -e "${GREEN}  ✓ Done${NC}"
-    else
-        echo -e "${RED}  ✗ Failed${NC}"
-        return 1
-    fi
-}
-
-# Record demos in logical order
-echo -e "${YELLOW}=== Phase 1: Basic demos ===${NC}"
-record_tape "$SCRIPT_DIR/install.tape"
-record_tape "$SCRIPT_DIR/quickstart.tape"
-record_tape "$SCRIPT_DIR/logs.tape"
-
-echo -e "${YELLOW}=== Phase 2: Update demo ===${NC}"
-record_tape "$SCRIPT_DIR/update.tape"
-
-echo -e "${YELLOW}=== Phase 3: Migration demo ===${NC}"
-record_tape "$SCRIPT_DIR/migration.tape"
-git -C /opt/stacks checkout compose-farm.yaml  # Reset after migration
-
-echo -e "${YELLOW}=== Phase 4: Apply demo ===${NC}"
-record_tape "$SCRIPT_DIR/apply.tape"
-
-# Move GIFs and WebMs from temp location to repo
-echo ""
-echo -e "${BLUE}Moving recordings to repo...${NC}"
-mv "$TEMP_OUTPUT"/*.gif "$OUTPUT_DIR/" 2>/dev/null || true
-mv "$TEMP_OUTPUT"/*.webm "$OUTPUT_DIR/" 2>/dev/null || true
-rmdir "$TEMP_OUTPUT" 2>/dev/null || true
-rmdir "$(dirname "$TEMP_OUTPUT")" 2>/dev/null || true
-
-echo ""
-echo -e "${GREEN}Done!${NC} Recordings saved to $OUTPUT_DIR/"
-ls -la "$OUTPUT_DIR"/*.gif "$OUTPUT_DIR"/*.webm 2>/dev/null || echo "No recordings found (check for errors above)"

docs/demos/web/conftest.py

@@ -21,24 +21,37 @@ import uvicorn
 
 from compose_farm.config import Config as CFConfig
 from compose_farm.config import load_config
+from compose_farm.executor import (
+    get_container_compose_labels as _original_get_compose_labels,
+)
+from compose_farm.glances import ContainerStats
+from compose_farm.glances import fetch_container_stats as _original_fetch_container_stats
 from compose_farm.state import load_state as _original_load_state
-from compose_farm.web.app import create_app
 from compose_farm.web.cdn import CDN_ASSETS, ensure_vendor_cache
 
+# NOTE: Do NOT import create_app here - it must be imported AFTER patches are applied
+# to ensure the patched get_config is used by all route modules
+
 if TYPE_CHECKING:
     from collections.abc import Generator
 
     from playwright.sync_api import BrowserContext, Page, Route
 
-# Stacks to exclude from demo recordings (exact match)
-DEMO_EXCLUDE_STACKS = {"arr"}
+# Substrings to exclude from demo recordings (case-insensitive)
+DEMO_EXCLUDE_PATTERNS = {"arr", "vpn", "tash"}
+
+
+def _should_exclude(name: str) -> bool:
+    """Check if a stack/container name should be excluded from demo."""
+    name_lower = name.lower()
+    return any(pattern in name_lower for pattern in DEMO_EXCLUDE_PATTERNS)
 
 
 def _get_filtered_config() -> CFConfig:
     """Load config but filter out excluded stacks."""
     config = load_config()
     filtered_stacks = {
-        name: host for name, host in config.stacks.items() if name not in DEMO_EXCLUDE_STACKS
+        name: host for name, host in config.stacks.items() if not _should_exclude(name)
     }
     return CFConfig(
         compose_dir=config.compose_dir,
@@ -46,6 +59,7 @@ def _get_filtered_config() -> CFConfig:
         stacks=filtered_stacks,
         traefik_file=config.traefik_file,
         traefik_stack=config.traefik_stack,
+        glances_stack=config.glances_stack,
         config_path=config.config_path,
     )
 
@@ -53,7 +67,37 @@ def _get_filtered_config() -> CFConfig:
 def _get_filtered_state(config: CFConfig) -> dict[str, str | list[str]]:
     """Load state but filter out excluded stacks."""
     state = _original_load_state(config)
-    return {name: host for name, host in state.items() if name not in DEMO_EXCLUDE_STACKS}
+    return {name: host for name, host in state.items() if not _should_exclude(name)}
+
+
+async def _filtered_fetch_container_stats(
+    host_name: str,
+    host_address: str,
+    port: int = 61208,
+    request_timeout: float = 10.0,
+) -> tuple[list[ContainerStats] | None, str | None]:
+    """Fetch container stats but filter out excluded containers."""
+    containers, error = await _original_fetch_container_stats(
+        host_name, host_address, port, request_timeout
+    )
+    if containers:
+        # Filter by container name (stack is empty at this point)
+        containers = [c for c in containers if not _should_exclude(c.name)]
+    return containers, error
+
+
+async def _filtered_get_compose_labels(
+    config: CFConfig,
+    host_name: str,
+) -> dict[str, tuple[str, str]]:
+    """Get compose labels but filter out excluded stacks."""
+    labels = await _original_get_compose_labels(config, host_name)
+    # Filter out containers whose stack (project) name should be excluded
+    return {
+        name: (stack, service)
+        for name, (stack, service) in labels.items()
+        if not _should_exclude(stack)
+    }
 
 
 @pytest.fixture(scope="session")
@@ -84,19 +128,23 @@ def server_url() -> Generator[str, None, None]:
 
     # Patch at source module level so all callers get filtered versions
     patches = [
-        # Patch load_state at source - all functions calling it get filtered state
+        # Patch load_config at source - get_config() calls this internally
+        patch("compose_farm.config.load_config", _get_filtered_config),
+        # Patch load_state at source and where imported
         patch("compose_farm.state.load_state", _get_filtered_state),
-        # Patch get_config where imported
-        patch("compose_farm.web.routes.pages.get_config", _get_filtered_config),
-        patch("compose_farm.web.routes.api.get_config", _get_filtered_config),
-        patch("compose_farm.web.routes.actions.get_config", _get_filtered_config),
-        patch("compose_farm.web.app.get_config", _get_filtered_config),
-        patch("compose_farm.web.ws.get_config", _get_filtered_config),
+        patch("compose_farm.web.routes.pages.load_state", _get_filtered_state),
+        # Patch container fetch to filter out excluded containers (Live Stats page)
+        patch("compose_farm.glances.fetch_container_stats", _filtered_fetch_container_stats),
+        # Patch compose labels to filter out excluded stacks
+        patch("compose_farm.executor.get_container_compose_labels", _filtered_get_compose_labels),
     ]
 
     for p in patches:
         p.start()
 
+    # Import create_app AFTER patches are started so route modules see patched get_config
+    from compose_farm.web.app import create_app  # noqa: PLC0415
+
     with socket.socket() as s:
         s.bind(("127.0.0.1", 0))
         port = s.getsockname()[1]
@@ -160,6 +208,7 @@ def recording_context(
             if url.startswith(url_prefix):
                 route.fulfill(status=200, content_type=content_type, body=filepath.read_bytes())
                 return
+        print(f"UNCACHED CDN request: {url}")
         route.abort("failed")
 
     context.route(re.compile(r"https://(cdn\.jsdelivr\.net|unpkg\.com)/.*"), handle_cdn)
@@ -176,6 +225,35 @@ def recording_page(recording_context: BrowserContext) -> Generator[Page, None, N
     page.close()
 
 
+@pytest.fixture
+def wide_recording_context(
+    browser: Any,  # pytest-playwright's browser fixture
+    recording_output_dir: Path,
+) -> Generator[BrowserContext, None, None]:
+    """Browser context with wider viewport for demos needing more horizontal space.
+
+    NOTE: This fixture does NOT use CDN interception (unlike recording_context).
+    CDN interception was causing inline scripts from containers.html to be
+    removed from the DOM, likely due to Tailwind's browser plugin behavior.
+    """
+    context = browser.new_context(
+        viewport={"width": 1920, "height": 1080},
+        record_video_dir=str(recording_output_dir),
+        record_video_size={"width": 1920, "height": 1080},
+    )
+
+    yield context
+    context.close()
+
+
+@pytest.fixture
+def wide_recording_page(wide_recording_context: BrowserContext) -> Generator[Page, None, None]:
+    """Page with wider viewport for demos needing more horizontal space."""
+    page = wide_recording_context.new_page()
+    yield page
+    page.close()
+
+
 # Demo helper functions
 
 

docs/demos/web/demo_console.py

@@ -60,10 +60,14 @@ def test_demo_console(recording_page: Page, server_url: str) -> None:
     page.keyboard.press("Enter")
     pause(page, 2500)  # Wait for output
 
-    # Scroll down to show the Editor section with Compose Farm config
-    editor_section = page.locator(".collapse", has_text="Editor").first
-    editor_section.scroll_into_view_if_needed()
-    pause(page, 800)
+    # Smoothly scroll down to show the Editor section with Compose Farm config
+    page.evaluate("""
+        const editor = document.getElementById('console-editor');
+        if (editor) {
+            editor.scrollIntoView({ behavior: 'smooth', block: 'center' });
+        }
+    """)
+    pause(page, 1200)  # Wait for smooth scroll animation
 
     # Wait for Monaco editor to load with config content
     page.wait_for_selector("#console-editor .monaco-editor", timeout=10000)

docs/demos/web/demo_live_stats.py

@@ -0,0 +1,85 @@
+"""Demo: Live Stats page.
+
+Records a ~20 second demo showing:
+- Navigating to Live Stats via command palette
+- Container table with real-time stats
+- Filtering containers
+- Sorting by different columns
+- Auto-refresh countdown
+
+Run: pytest docs/demos/web/demo_live_stats.py -v --no-cov
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+import pytest
+from conftest import (
+    open_command_palette,
+    pause,
+    slow_type,
+    wait_for_sidebar,
+)
+
+if TYPE_CHECKING:
+    from playwright.sync_api import Page
+
+
+@pytest.mark.browser  # type: ignore[misc]
+def test_demo_live_stats(wide_recording_page: Page, server_url: str) -> None:
+    """Record Live Stats page demo."""
+    page = wide_recording_page
+
+    # Start on dashboard
+    page.goto(server_url)
+    wait_for_sidebar(page)
+    pause(page, 1000)
+
+    # Navigate to Live Stats via command palette
+    open_command_palette(page)
+    pause(page, 400)
+    slow_type(page, "#cmd-input", "live", delay=100)
+    pause(page, 500)
+    page.keyboard.press("Enter")
+    page.wait_for_url("**/live-stats", timeout=5000)
+
+    # Wait for containers to load (may take ~10s on first load due to SSH)
+    page.wait_for_selector("#container-rows tr:not(:has(.loading))", timeout=30000)
+    pause(page, 2000)  # Let viewer see the full table with timer
+
+    # Demonstrate filtering
+    slow_type(page, "#filter-input", "grocy", delay=100)
+    pause(page, 1500)  # Show filtered results
+
+    # Clear filter
+    page.fill("#filter-input", "")
+    pause(page, 1000)
+
+    # Sort by memory (click header)
+    page.click("th:has-text('Mem')")
+    pause(page, 1500)
+
+    # Sort by CPU
+    page.click("th:has-text('CPU')")
+    pause(page, 1500)
+
+    # Sort by host
+    page.click("th:has-text('Host')")
+    pause(page, 1500)
+
+    # Watch auto-refresh timer count down
+    pause(page, 3500)  # Wait for refresh to happen
+
+    # Hover on action menu to show pause behavior
+    action_btn = page.locator('button[onclick^="openActionMenu"]').first
+    action_btn.scroll_into_view_if_needed()
+    action_btn.hover()
+    pause(page, 2000)  # Show paused state (timer shows ⏸) and action menu
+
+    # Move away to close menu and resume refresh
+    page.locator("h2").first.hover()  # Move to header
+    pause(page, 3500)  # Watch countdown resume and refresh happen
+
+    # Final pause
+    pause(page, 1000)

docs/demos/web/demo_shell.py

@@ -1,9 +1,11 @@
-"""Demo: Container shell exec.
+"""Demo: Container shell exec via command palette.
 
-Records a ~25 second demo showing:
-- Navigating to a stack page
-- Clicking Shell button on a container
-- Running top command inside the container
+Records a ~35 second demo showing:
+- Navigating to immich stack (multiple containers)
+- Using command palette with fuzzy matching ("sh mach") to open shell
+- Running a command
+- Using command palette to switch to server container shell
+- Running another command
 
 Run: pytest docs/demos/web/demo_shell.py -v --no-cov
 """
@@ -14,6 +16,7 @@ from typing import TYPE_CHECKING
 
 import pytest
 from conftest import (
+    open_command_palette,
     pause,
     slow_type,
     wait_for_sidebar,
@@ -33,39 +36,71 @@ def test_demo_shell(recording_page: Page, server_url: str) -> None:
     wait_for_sidebar(page)
     pause(page, 800)
 
-    # Navigate to a stack with a running container (grocy)
-    page.locator("#sidebar-stacks a", has_text="grocy").click()
-    page.wait_for_url("**/stack/grocy", timeout=5000)
+    # Navigate to immich via command palette (has multiple containers)
+    open_command_palette(page)
+    pause(page, 400)
+    slow_type(page, "#cmd-input", "immich", delay=100)
+    pause(page, 600)
+    page.keyboard.press("Enter")
+    page.wait_for_url("**/stack/immich", timeout=5000)
     pause(page, 1500)
 
-    # Wait for containers list to load (loaded via HTMX)
+    # Wait for containers list to load (so shell commands are available)
     page.wait_for_selector("#containers-list button", timeout=10000)
     pause(page, 800)
 
-    # Click Shell button on the first container
-    shell_btn = page.locator("#containers-list button", has_text="Shell").first
-    shell_btn.click()
+    # Use command palette with fuzzy matching: "sh mach" -> "Shell: immich-machine-learning"
+    open_command_palette(page)
+    pause(page, 400)
+    slow_type(page, "#cmd-input", "sh mach", delay=100)
+    pause(page, 600)
+    page.keyboard.press("Enter")
     pause(page, 1000)
 
     # Wait for exec terminal to appear
     page.wait_for_selector("#exec-terminal .xterm", timeout=10000)
 
-    # Scroll down to make the terminal visible
-    page.locator("#exec-terminal").scroll_into_view_if_needed()
-    pause(page, 2000)
+    # Smoothly scroll down to make the terminal visible
+    page.evaluate("""
+        const terminal = document.getElementById('exec-terminal');
+        if (terminal) {
+            terminal.scrollIntoView({ behavior: 'smooth', block: 'center' });
+        }
+    """)
+    pause(page, 1200)
 
-    # Run top command
-    slow_type(page, "#exec-terminal .xterm-helper-textarea", "top", delay=100)
+    # Run python version command
+    slow_type(page, "#exec-terminal .xterm-helper-textarea", "python3 --version", delay=60)
     pause(page, 300)
     page.keyboard.press("Enter")
-    pause(page, 4000)  # Let top run for a bit
+    pause(page, 1500)
 
-    # Press q to quit top
-    page.keyboard.press("q")
+    # Blur the terminal to release focus (won't scroll)
+    page.evaluate("document.activeElement?.blur()")
+    pause(page, 500)
+
+    # Use command palette to switch to server container: "sh serv" -> "Shell: immich-server"
+    open_command_palette(page)
+    pause(page, 400)
+    slow_type(page, "#cmd-input", "sh serv", delay=100)
+    pause(page, 600)
+    page.keyboard.press("Enter")
     pause(page, 1000)
 
-    # Run another command to show it's interactive
-    slow_type(page, "#exec-terminal .xterm-helper-textarea", "ps aux | head", delay=60)
+    # Wait for new terminal
+    page.wait_for_selector("#exec-terminal .xterm", timeout=10000)
+
+    # Scroll to terminal
+    page.evaluate("""
+        const terminal = document.getElementById('exec-terminal');
+        if (terminal) {
+            terminal.scrollIntoView({ behavior: 'smooth', block: 'center' });
+        }
+    """)
+    pause(page, 1200)
+
+    # Run ls command
+    slow_type(page, "#exec-terminal .xterm-helper-textarea", "ls /usr/src/app", delay=60)
     pause(page, 300)
     page.keyboard.press("Enter")
     pause(page, 2000)

docs/demos/web/demo_stack.py

@@ -55,9 +55,14 @@ def test_demo_stack(recording_page: Page, server_url: str) -> None:
     page.wait_for_selector("#compose-editor .monaco-editor", timeout=10000)
     pause(page, 2000)  # Let viewer see the compose file
 
-    # Scroll down slightly to show more of the editor
-    page.locator("#compose-editor").scroll_into_view_if_needed()
-    pause(page, 1500)
+    # Smoothly scroll down to show more of the editor
+    page.evaluate("""
+        const editor = document.getElementById('compose-editor');
+        if (editor) {
+            editor.scrollIntoView({ behavior: 'smooth', block: 'center' });
+        }
+    """)
+    pause(page, 1200)  # Wait for smooth scroll animation
 
     # Close the compose file section
     compose_collapse.locator("input[type=checkbox]").click(force=True)

docs/demos/web/demo_workflow.py

@@ -5,7 +5,7 @@ Records a comprehensive demo (~60 seconds) combining all major features:
 2. Editor showing Compose Farm YAML config
 3. Command palette navigation to grocy stack
 4. Stack actions: up, logs
-5. Switch to mealie stack via command palette, run update
+5. Switch to dozzle stack via command palette, run update
 6. Dashboard overview
 7. Theme cycling via command palette
 
@@ -126,13 +126,13 @@ def _demo_stack_actions(page: Page) -> None:
     page.wait_for_selector("#terminal-output .xterm", timeout=5000)
     pause(page, 2500)
 
-    # Switch to mealie via command palette
+    # Switch to dozzle via command palette (on nas for lower latency)
     open_command_palette(page)
     pause(page, 300)
-    slow_type(page, "#cmd-input", "mealie", delay=100)
+    slow_type(page, "#cmd-input", "dozzle", delay=100)
     pause(page, 400)
     page.keyboard.press("Enter")
-    page.wait_for_url("**/stack/mealie", timeout=5000)
+    page.wait_for_url("**/stack/dozzle", timeout=5000)
     pause(page, 1000)
 
     # Run update action
@@ -162,32 +162,20 @@ def _demo_dashboard_and_themes(page: Page, server_url: str) -> None:
     page.evaluate("window.scrollTo(0, 0)")
     pause(page, 600)
 
-    # Open theme picker and arrow down to Luxury (shows live preview)
-    # Theme order: light, dark, cupcake, bumblebee, emerald, corporate, synthwave,
-    # retro, cyberpunk, valentine, halloween, garden, forest, aqua, lofi, pastel,
-    # fantasy, wireframe, black, luxury (index 19)
+    # Open theme picker and arrow down to Dracula (shows live preview)
     page.locator("#theme-btn").click()
     page.wait_for_selector("#cmd-palette[open]", timeout=2000)
     pause(page, 400)
 
-    # Arrow down through themes with live preview until we reach Luxury
+    # Arrow down through themes with live preview until we reach Dracula
     for _ in range(19):
         page.keyboard.press("ArrowDown")
         pause(page, 180)
 
-    # Select Luxury theme
+    # Select Dracula theme and end on it
     pause(page, 400)
     page.keyboard.press("Enter")
-    pause(page, 1000)
-
-    # Return to dark theme
-    page.locator("#theme-btn").click()
-    page.wait_for_selector("#cmd-palette[open]", timeout=2000)
-    pause(page, 300)
-    slow_type(page, "#cmd-input", " dark", delay=80)
-    pause(page, 400)
-    page.keyboard.press("Enter")
-    pause(page, 1000)
+    pause(page, 1500)
 
 
 @pytest.mark.browser  # type: ignore[misc]

docs/demos/web/record.py

@@ -37,6 +37,7 @@ DEMOS = [
     "workflow",
     "console",
     "shell",
+    "live_stats",
 ]
 
 # High-quality ffmpeg settings for VP8 encoding
@@ -88,9 +89,9 @@ def patch_playwright_video_quality() -> None:
     console.print("[green]Patched Playwright for high-quality video recording[/green]")
 
 
-def record_demo(name: str) -> Path | None:
+def record_demo(name: str, index: int, total: int) -> Path | None:
     """Run a single demo and return the video path."""
-    console.print(f"[green]Recording:[/green] web-{name}")
+    console.print(f"[cyan][{index}/{total}][/cyan] [green]Recording:[/green] web-{name}")
 
     demo_file = SCRIPT_DIR / f"demo_{name}.py"
     if not demo_file.exists():
@@ -227,9 +228,7 @@ def main() -> int:
 
     try:
         for i, demo in enumerate(demos_to_record, 1):
-            console.print(f"[yellow]=== Demo {i}/{len(demos_to_record)}: {demo} ===[/yellow]")
-
-            video_path = record_demo(demo)
+            video_path = record_demo(demo, i, len(demos_to_record))
             if video_path:
                 webm, gif = move_recording(video_path, demo)
                 results[demo] = (webm, gif)

docs/getting-started.md

@@ -54,6 +54,25 @@ docker run --rm \
   ghcr.io/basnijholt/compose-farm up --all
 ```
 
+**Running as non-root user** (recommended for NFS mounts):
+
+By default, containers run as root. To preserve file ownership on mounted volumes, set these environment variables in your `.env` file:
+
+```bash
+# Add to .env file (one-time setup)
+echo "CF_UID=$(id -u)" >> .env
+echo "CF_GID=$(id -g)" >> .env
+echo "CF_HOME=$HOME" >> .env
+echo "CF_USER=$USER" >> .env
+```
+
+Or use [direnv](https://direnv.net/) to auto-set these variables when entering the directory:
+```bash
+cp .envrc.example .envrc && direnv allow
+```
+
+This ensures files like `compose-farm-state.yaml` and web UI edits are owned by your user instead of root. The `CF_USER` variable is required for SSH to work when running as a non-root user.
+
 ### Verify Installation
 
 ```bash

docs/web-ui.md

@@ -51,10 +51,32 @@ Press `Ctrl+K` (or `Cmd+K` on macOS) to open the command palette. Use fuzzy sear
 ### Dashboard (`/`)
 
 - Stack overview with status indicators
-- Host statistics
+- Host statistics (CPU, memory, disk, load via Glances)
 - Pending operations (migrations, orphaned stacks)
 - Quick actions via command palette
 
+### Live Stats (`/live-stats`)
+
+Real-time container monitoring across all hosts, powered by [Glances](https://nicolargo.github.io/glances/).
+
+- **Live metrics**: CPU, memory, network I/O for every container
+- **Auto-refresh**: Updates every 3 seconds (pauses when dropdown menus are open)
+- **Filtering**: Type to filter containers by name, stack, host, or image
+- **Sorting**: Click column headers to sort by any metric
+- **Update detection**: Shows when container images have updates available
+
+<video autoplay loop muted playsinline>
+  <source src="/assets/web-live_stats.webm" type="video/webm">
+</video>
+
+#### Requirements
+
+Live Stats requires Glances to be deployed on all hosts:
+
+1. Add `glances_stack: glances` to your `compose-farm.yaml`
+2. Deploy a Glances stack that runs on all hosts (see [example](https://github.com/basnijholt/compose-farm/tree/main/examples/glances))
+3. Glances must expose its REST API on port 61208
+
 ### Stack Detail (`/stack/{name}`)
 
 - Compose file editor (Monaco)
@@ -63,6 +85,8 @@ Press `Ctrl+K` (or `Cmd+K` on macOS) to open the command palette. Use fuzzy sear
 - Container shell access (exec into running containers)
 - Terminal output for running commands
 
+Files are automatically backed up before saving to `~/.config/compose-farm/backups/`.
+
 ### Console (`/console`)
 
 - Full shell access to any host

justfile

@@ -45,6 +45,14 @@ doc:
 kill-doc:
     lsof -ti :9002 | xargs kill -9 2>/dev/null || true
 
+# Record CLI demos (all or specific: just record-cli quickstart)
+record-cli *demos:
+    python docs/demos/cli/record.py {{demos}}
+
+# Record web UI demos (all or specific: just record-web navigation)
+record-web *demos:
+    python docs/demos/web/record.py {{demos}}
+
 # Clean up build artifacts and caches
 clean:
     rm -rf .pytest_cache .mypy_cache .ruff_cache .coverage htmlcov dist build

pyproject.toml

@@ -53,6 +53,7 @@ web = [
     "fastapi[standard]>=0.109.0",
     "jinja2>=3.1.0",
     "websockets>=12.0",
+    "humanize>=4.0.0",
 ]
 
 [project.urls]

src/compose_farm/cli/lifecycle.py

@@ -3,10 +3,13 @@
 from __future__ import annotations
 
 from pathlib import Path
-from typing import Annotated
+from typing import TYPE_CHECKING, Annotated
 
 import typer
 
+if TYPE_CHECKING:
+    from compose_farm.config import Config
+
 from compose_farm.cli.app import app
 from compose_farm.cli.common import (
     AllOption,
@@ -23,9 +26,14 @@ from compose_farm.cli.common import (
     validate_host_for_stack,
     validate_stacks,
 )
+from compose_farm.cli.management import _discover_stacks_full
 from compose_farm.console import MSG_DRY_RUN, console, print_error, print_success
 from compose_farm.executor import run_compose_on_host, run_on_stacks, run_sequential_on_stacks
-from compose_farm.operations import stop_orphaned_stacks, up_stacks
+from compose_farm.operations import (
+    stop_orphaned_stacks,
+    stop_stray_stacks,
+    up_stacks,
+)
 from compose_farm.state import (
     get_orphaned_stacks,
     get_stack_host,
@@ -208,8 +216,23 @@ def update(
     report_results(results)
 
 
+def _discover_strays(cfg: Config) -> dict[str, list[str]]:
+    """Discover stacks running on unauthorized hosts by scanning all hosts."""
+    _, strays, duplicates = _discover_stacks_full(cfg)
+
+    # Merge duplicates into strays (for single-host stacks on multiple hosts,
+    # keep correct host and stop others)
+    for stack, running_hosts in duplicates.items():
+        configured = cfg.get_hosts(stack)[0]
+        stray_hosts = [h for h in running_hosts if h != configured]
+        if stray_hosts:
+            strays[stack] = stray_hosts
+
+    return strays
+
+
 @app.command(rich_help_panel="Lifecycle")
-def apply(  # noqa: PLR0912 (multi-phase reconciliation needs these branches)
+def apply(  # noqa: C901, PLR0912, PLR0915 (multi-phase reconciliation needs these branches)
     dry_run: Annotated[
         bool,
         typer.Option("--dry-run", "-n", help="Show what would change without executing"),
@@ -218,23 +241,29 @@ def apply(  # noqa: PLR0912 (multi-phase reconciliation needs these branches)
         bool,
         typer.Option("--no-orphans", help="Only migrate, don't stop orphaned stacks"),
     ] = False,
+    no_strays: Annotated[
+        bool,
+        typer.Option("--no-strays", help="Don't stop stray stacks (running on wrong host)"),
+    ] = False,
     full: Annotated[
         bool,
         typer.Option("--full", "-f", help="Also run up on all stacks to apply config changes"),
     ] = False,
     config: ConfigOption = None,
 ) -> None:
-    """Make reality match config (start, migrate, stop as needed).
+    """Make reality match config (start, migrate, stop strays/orphans as needed).
 
     This is the "reconcile" command that ensures running stacks match your
     config file. It will:
 
     1. Stop orphaned stacks (in state but removed from config)
-    2. Migrate stacks on wrong host (host in state ≠ host in config)
-    3. Start missing stacks (in config but not in state)
+    2. Stop stray stacks (running on unauthorized hosts)
+    3. Migrate stacks on wrong host (host in state ≠ host in config)
+    4. Start missing stacks (in config but not in state)
 
     Use --dry-run to preview changes before applying.
-    Use --no-orphans to only migrate/start without stopping orphaned stacks.
+    Use --no-orphans to skip stopping orphaned stacks.
+    Use --no-strays to skip stopping stray stacks.
     Use --full to also run 'up' on all stacks (picks up compose/env changes).
     """
     cfg = load_config_or_exit(config)
@@ -242,16 +271,28 @@ def apply(  # noqa: PLR0912 (multi-phase reconciliation needs these branches)
     migrations = get_stacks_needing_migration(cfg)
     missing = get_stacks_not_in_state(cfg)
 
+    strays: dict[str, list[str]] = {}
+    if not no_strays:
+        console.print("[dim]Scanning hosts for stray containers...[/]")
+        strays = _discover_strays(cfg)
+
     # For --full: refresh all stacks not already being started/migrated
     handled = set(migrations) | set(missing)
     to_refresh = [stack for stack in cfg.stacks if stack not in handled] if full else []
 
     has_orphans = bool(orphaned) and not no_orphans
+    has_strays = bool(strays)
     has_migrations = bool(migrations)
     has_missing = bool(missing)
     has_refresh = bool(to_refresh)
 
-    if not has_orphans and not has_migrations and not has_missing and not has_refresh:
+    if (
+        not has_orphans
+        and not has_strays
+        and not has_migrations
+        and not has_missing
+        and not has_refresh
+    ):
         print_success("Nothing to apply - reality matches config")
         return
 
@@ -260,6 +301,14 @@ def apply(  # noqa: PLR0912 (multi-phase reconciliation needs these branches)
         console.print(f"[yellow]Orphaned stacks to stop ({len(orphaned)}):[/]")
         for svc, hosts in orphaned.items():
             console.print(f"  [cyan]{svc}[/] on [magenta]{format_host(hosts)}[/]")
+    if has_strays:
+        console.print(f"[red]Stray stacks to stop ({len(strays)}):[/]")
+        for stack, hosts in strays.items():
+            configured = cfg.get_hosts(stack)
+            console.print(
+                f"  [cyan]{stack}[/] on [magenta]{', '.join(hosts)}[/] "
+                f"[dim](should be on {', '.join(configured)})[/]"
+            )
     if has_migrations:
         console.print(f"[cyan]Stacks to migrate ({len(migrations)}):[/]")
         for stack in migrations:
@@ -288,21 +337,26 @@ def apply(  # noqa: PLR0912 (multi-phase reconciliation needs these branches)
         console.print("[yellow]Stopping orphaned stacks...[/]")
         all_results.extend(run_async(stop_orphaned_stacks(cfg)))
 
-    # 2. Migrate stacks on wrong host
+    # 2. Stop stray stacks (running on unauthorized hosts)
+    if has_strays:
+        console.print("[red]Stopping stray stacks...[/]")
+        all_results.extend(run_async(stop_stray_stacks(cfg, strays)))
+
+    # 3. Migrate stacks on wrong host
     if has_migrations:
         console.print("[cyan]Migrating stacks...[/]")
         migrate_results = run_async(up_stacks(cfg, migrations, raw=True))
         all_results.extend(migrate_results)
         maybe_regenerate_traefik(cfg, migrate_results)
 
-    # 3. Start missing stacks (reuse up_stacks which handles state updates)
+    # 4. Start missing stacks (reuse up_stacks which handles state updates)
     if has_missing:
         console.print("[green]Starting missing stacks...[/]")
         start_results = run_async(up_stacks(cfg, missing, raw=True))
         all_results.extend(start_results)
         maybe_regenerate_traefik(cfg, start_results)
 
-    # 4. Refresh remaining stacks (--full: run up to apply config changes)
+    # 5. Refresh remaining stacks (--full: run up to apply config changes)
     if has_refresh:
         console.print("[blue]Refreshing stacks...[/]")
         refresh_results = run_async(up_stacks(cfg, to_refresh, raw=True))

src/compose_farm/cli/management.py

@@ -37,22 +37,23 @@ from compose_farm.console import (
 )
 from compose_farm.executor import (
     CommandResult,
+    get_running_stacks_on_host,
     is_local,
     run_command,
 )
 from compose_farm.logs import (
     DEFAULT_LOG_PATH,
     SnapshotEntry,
-    collect_stack_entries,
+    collect_stacks_entries_on_host,
     isoformat,
     load_existing_entries,
     merge_entries,
     write_toml,
 )
 from compose_farm.operations import (
+    build_discovery_results,
     check_host_compatibility,
     check_stack_requirements,
-    discover_stack_host,
 )
 from compose_farm.state import get_orphaned_stacks, load_state, save_state
 from compose_farm.traefik import generate_traefik_config, render_traefik_config
@@ -60,38 +61,39 @@ from compose_farm.traefik import generate_traefik_config, render_traefik_config
 # --- Sync helpers ---
 
 
-def _discover_stacks(cfg: Config, stacks: list[str] | None = None) -> dict[str, str | list[str]]:
-    """Discover running stacks with a progress bar."""
-    stack_list = stacks if stacks is not None else list(cfg.stacks)
-    results = run_parallel_with_progress(
-        "Discovering",
-        stack_list,
-        lambda s: discover_stack_host(cfg, s),
-    )
-    return {svc: host for svc, host in results if host is not None}
-
-
 def _snapshot_stacks(
     cfg: Config,
-    stacks: list[str],
+    discovered: dict[str, str | list[str]],
     log_path: Path | None,
 ) -> Path:
-    """Capture image digests with a progress bar."""
+    """Capture image digests using batched SSH calls (1 per host).
+
+    Args:
+        cfg: Configuration
+        discovered: Dict mapping stack -> host(s) where it's running
+        log_path: Optional path to write the log file
+
+    Returns:
+        Path to the written log file.
+
+    """
     effective_log_path = log_path or DEFAULT_LOG_PATH
     now_dt = datetime.now(UTC)
     now_iso = isoformat(now_dt)
 
-    async def collect_stack(stack: str) -> tuple[str, list[SnapshotEntry]]:
-        try:
-            return stack, await collect_stack_entries(cfg, stack, now=now_dt)
-        except RuntimeError:
-            return stack, []
+    # Group stacks by host for batched SSH calls
+    stacks_by_host: dict[str, set[str]] = {}
+    for stack, hosts in discovered.items():
+        # Use first host for multi-host stacks (they use the same images)
+        host = hosts[0] if isinstance(hosts, list) else hosts
+        stacks_by_host.setdefault(host, set()).add(stack)
 
-    results = run_parallel_with_progress(
-        "Capturing",
-        stacks,
-        collect_stack,
-    )
+    # Collect entries with 1 SSH call per host (with progress bar)
+    async def collect_on_host(host: str) -> tuple[str, list[SnapshotEntry]]:
+        entries = await collect_stacks_entries_on_host(cfg, host, stacks_by_host[host], now=now_dt)
+        return host, entries
+
+    results = run_parallel_with_progress("Capturing", list(stacks_by_host.keys()), collect_on_host)
     snapshot_entries = [entry for _, entries in results for entry in entries]
 
     if not snapshot_entries:
@@ -147,6 +149,61 @@ def _report_sync_changes(
             console.print(f"  [red]-[/] [cyan]{stack}[/] (was on [magenta]{host_str}[/])")
 
 
+def _discover_stacks_full(
+    cfg: Config,
+    stacks: list[str] | None = None,
+) -> tuple[dict[str, str | list[str]], dict[str, list[str]], dict[str, list[str]]]:
+    """Discover running stacks with full host scanning for stray detection.
+
+    Queries each host once for all running stacks (with progress bar),
+    then delegates to build_discovery_results for categorization.
+    """
+    all_hosts = list(cfg.hosts.keys())
+
+    # Query each host for running stacks (with progress bar)
+    async def get_stacks_on_host(host: str) -> tuple[str, set[str]]:
+        running = await get_running_stacks_on_host(cfg, host)
+        return host, running
+
+    host_results = run_parallel_with_progress("Discovering", all_hosts, get_stacks_on_host)
+    running_on_host: dict[str, set[str]] = dict(host_results)
+
+    return build_discovery_results(cfg, running_on_host, stacks)
+
+
+def _report_stray_stacks(
+    strays: dict[str, list[str]],
+    cfg: Config,
+) -> None:
+    """Report stacks running on unauthorized hosts."""
+    if strays:
+        console.print(f"\n[red]Stray stacks[/] (running on wrong host, {len(strays)}):")
+        console.print("[dim]Run [bold]cf apply[/bold] to stop them.[/]")
+        for stack in sorted(strays):
+            stray_hosts = strays[stack]
+            configured = cfg.get_hosts(stack)
+            console.print(
+                f"  [red]![/] [cyan]{stack}[/] on [magenta]{', '.join(stray_hosts)}[/] "
+                f"[dim](should be on {', '.join(configured)})[/]"
+            )
+
+
+def _report_duplicate_stacks(duplicates: dict[str, list[str]], cfg: Config) -> None:
+    """Report single-host stacks running on multiple hosts."""
+    if duplicates:
+        console.print(
+            f"\n[yellow]Duplicate stacks[/] (running on multiple hosts, {len(duplicates)}):"
+        )
+        console.print("[dim]Run [bold]cf apply[/bold] to stop extras.[/]")
+        for stack in sorted(duplicates):
+            hosts = duplicates[stack]
+            configured = cfg.get_hosts(stack)[0]
+            console.print(
+                f"  [yellow]![/] [cyan]{stack}[/] on [magenta]{', '.join(hosts)}[/] "
+                f"[dim](should only be on {configured})[/]"
+            )
+
+
 # --- Check helpers ---
 
 
@@ -440,7 +497,7 @@ def refresh(
 
     current_state = load_state(cfg)
 
-    discovered = _discover_stacks(cfg, stack_list)
+    discovered, strays, duplicates = _discover_stacks_full(cfg, stack_list)
 
     # Calculate changes (only for the stacks we're refreshing)
     added = [s for s in discovered if s not in current_state]
@@ -463,6 +520,9 @@ def refresh(
     else:
         print_success("State is already in sync.")
 
+    _report_stray_stacks(strays, cfg)
+    _report_duplicate_stacks(duplicates, cfg)
+
     if dry_run:
         console.print(f"\n{MSG_DRY_RUN}")
         return
@@ -475,10 +535,10 @@ def refresh(
         save_state(cfg, new_state)
         print_success(f"State updated: {len(new_state)} stacks tracked.")
 
-    # Capture image digests for running stacks
+    # Capture image digests for running stacks (1 SSH call per host)
     if discovered:
         try:
-            path = _snapshot_stacks(cfg, list(discovered.keys()), log_path)
+            path = _snapshot_stacks(cfg, discovered, log_path)
             print_success(f"Digests written to {path}")
         except RuntimeError as exc:
             print_warning(str(exc))

src/compose_farm/config.py

@@ -31,6 +31,9 @@ class Config(BaseModel, extra="forbid"):
     stacks: dict[str, str | list[str]]  # stack_name -> host_name or list of hosts
     traefik_file: Path | None = None  # Auto-regenerate traefik config after up/down
     traefik_stack: str | None = None  # Stack name for Traefik (skip its host in file-provider)
+    glances_stack: str | None = (
+        None  # Stack name for Glances (enables host resource stats in web UI)
+    )
     config_path: Path = Path()  # Set by load_config()
 
     def get_state_path(self) -> Path:

src/compose_farm/executor.py

@@ -5,6 +5,7 @@ from __future__ import annotations
 import asyncio
 import socket
 import subprocess
+import time
 from dataclasses import dataclass
 from functools import lru_cache
 from typing import TYPE_CHECKING, Any
@@ -23,6 +24,38 @@ LOCAL_ADDRESSES = frozenset({"local", "localhost", "127.0.0.1", "::1"})
 _DEFAULT_SSH_PORT = 22
 
 
+class TTLCache:
+    """Simple TTL cache for async function results."""
+
+    def __init__(self, ttl_seconds: float = 30.0) -> None:
+        """Initialize cache with default TTL in seconds."""
+        # Cache stores: key -> (timestamp, value, item_ttl)
+        self._cache: dict[str, tuple[float, Any, float]] = {}
+        self._default_ttl = ttl_seconds
+
+    def get(self, key: str) -> Any | None:
+        """Get value if exists and not expired."""
+        if key in self._cache:
+            timestamp, value, item_ttl = self._cache[key]
+            if time.monotonic() - timestamp < item_ttl:
+                return value
+            del self._cache[key]
+        return None
+
+    def set(self, key: str, value: Any, ttl_seconds: float | None = None) -> None:
+        """Set value with current timestamp and optional custom TTL."""
+        ttl = ttl_seconds if ttl_seconds is not None else self._default_ttl
+        self._cache[key] = (time.monotonic(), value, ttl)
+
+    def clear(self) -> None:
+        """Clear all cached values."""
+        self._cache.clear()
+
+
+# Cache compose labels per host for 30 seconds
+_compose_labels_cache = TTLCache(ttl_seconds=30.0)
+
+
 def _print_compose_command(
     host_name: str,
     compose_dir: str,
@@ -158,15 +191,20 @@ def ssh_connect_kwargs(host: Host) -> dict[str, Any]:
         "port": host.port,
         "username": host.user,
         "known_hosts": None,
+        "gss_auth": False,  # Disable GSSAPI - causes multi-second delays
     }
-    # Add SSH agent path (auto-detect forwarded agent if needed)
-    agent_path = get_ssh_auth_sock()
-    if agent_path:
-        kwargs["agent_path"] = agent_path
-    # Add key file fallback for when SSH agent is unavailable
+    # Add key file fallback (prioritized over agent if present)
     key_path = get_key_path()
+    agent_path = get_ssh_auth_sock()
+
     if key_path:
+        # If dedicated key exists, force use of it and ignore agent
+        # This avoids issues with stale/broken forwarded agents in Docker
         kwargs["client_keys"] = [str(key_path)]
+    elif agent_path:
+        # Fallback to agent if no dedicated key
+        kwargs["agent_path"] = agent_path
+
     return kwargs
 
 
@@ -497,6 +535,72 @@ async def check_stack_running(
     return result.success and bool(result.stdout.strip())
 
 
+async def get_running_stacks_on_host(
+    config: Config,
+    host_name: str,
+) -> set[str]:
+    """Get all running compose stacks on a host in a single SSH call.
+
+    Uses docker ps with the compose.project label to identify running stacks.
+    Much more efficient than checking each stack individually.
+    """
+    host = config.hosts[host_name]
+
+    # Get unique project names from running containers
+    command = "docker ps --format '{{.Label \"com.docker.compose.project\"}}' | sort -u"
+    result = await run_command(host, command, stack=host_name, stream=False, prefix="")
+
+    if not result.success:
+        return set()
+
+    # Filter out empty lines and return as set
+    return {line.strip() for line in result.stdout.splitlines() if line.strip()}
+
+
+async def get_container_compose_labels(
+    config: Config,
+    host_name: str,
+) -> dict[str, tuple[str, str]]:
+    """Get compose labels for all containers on a host.
+
+    Returns dict of container_name -> (project, service).
+    Includes all containers (-a flag) since Glances shows stopped containers too.
+    Falls back to empty dict on timeout/error (5s timeout).
+    Results are cached for 30 seconds to reduce SSH overhead.
+    """
+    # Check cache first
+    cached: dict[str, tuple[str, str]] | None = _compose_labels_cache.get(host_name)
+    if cached is not None:
+        return cached
+
+    host = config.hosts[host_name]
+    cmd = (
+        "docker ps -a --format "
+        '\'{{.Names}}\t{{.Label "com.docker.compose.project"}}\t'
+        '{{.Label "com.docker.compose.service"}}\''
+    )
+
+    try:
+        async with asyncio.timeout(5.0):
+            result = await run_command(host, cmd, stack=host_name, stream=False, prefix="")
+    except TimeoutError:
+        return {}
+    except Exception:
+        return {}
+
+    labels: dict[str, tuple[str, str]] = {}
+    if result.success:
+        for line in result.stdout.splitlines():
+            parts = line.strip().split("\t")
+            if len(parts) >= 3:  # noqa: PLR2004
+                name, project, service = parts[0], parts[1], parts[2]
+                labels[name] = (project or "", service or "")
+
+    # Cache the result
+    _compose_labels_cache.set(host_name, labels)
+    return labels
+
+
 async def _batch_check_existence(
     config: Config,
     host_name: str,

src/compose_farm/glances.py

@@ -0,0 +1,236 @@
+"""Glances API client for host resource monitoring."""
+
+from __future__ import annotations
+
+import asyncio
+from dataclasses import dataclass
+from typing import TYPE_CHECKING, Any
+
+if TYPE_CHECKING:
+    from .config import Config
+
+# Default Glances REST API port
+DEFAULT_GLANCES_PORT = 61208
+
+
+@dataclass
+class HostStats:
+    """Resource statistics for a host."""
+
+    host: str
+    cpu_percent: float
+    mem_percent: float
+    swap_percent: float
+    load: float
+    disk_percent: float
+    net_rx_rate: float = 0.0  # bytes/sec
+    net_tx_rate: float = 0.0  # bytes/sec
+    error: str | None = None
+
+    @classmethod
+    def from_error(cls, host: str, error: str) -> HostStats:
+        """Create a HostStats with an error."""
+        return cls(
+            host=host,
+            cpu_percent=0,
+            mem_percent=0,
+            swap_percent=0,
+            load=0,
+            disk_percent=0,
+            net_rx_rate=0,
+            net_tx_rate=0,
+            error=error,
+        )
+
+
+async def fetch_host_stats(
+    host_name: str,
+    host_address: str,
+    port: int = DEFAULT_GLANCES_PORT,
+    request_timeout: float = 10.0,
+) -> HostStats:
+    """Fetch stats from a single host's Glances API."""
+    import httpx  # noqa: PLC0415
+
+    base_url = f"http://{host_address}:{port}/api/4"
+
+    try:
+        async with httpx.AsyncClient(timeout=request_timeout) as client:
+            # Fetch quicklook stats (CPU, mem, load)
+            response = await client.get(f"{base_url}/quicklook")
+            if not response.is_success:
+                return HostStats.from_error(host_name, f"HTTP {response.status_code}")
+            data = response.json()
+
+            # Fetch filesystem stats for disk usage (root fs or max across all)
+            disk_percent = 0.0
+            try:
+                fs_response = await client.get(f"{base_url}/fs")
+                if fs_response.is_success:
+                    fs_data = fs_response.json()
+                    root = next((fs for fs in fs_data if fs.get("mnt_point") == "/"), None)
+                    disk_percent = (
+                        root.get("percent", 0)
+                        if root
+                        else max((fs.get("percent", 0) for fs in fs_data), default=0)
+                    )
+            except httpx.HTTPError:
+                pass  # Disk stats are optional
+
+            # Fetch network stats for rate (sum across non-loopback interfaces)
+            net_rx_rate, net_tx_rate = 0.0, 0.0
+            try:
+                net_response = await client.get(f"{base_url}/network")
+                if net_response.is_success:
+                    for iface in net_response.json():
+                        if not iface.get("interface_name", "").startswith("lo"):
+                            net_rx_rate += iface.get("bytes_recv_rate_per_sec") or 0
+                            net_tx_rate += iface.get("bytes_sent_rate_per_sec") or 0
+            except httpx.HTTPError:
+                pass  # Network stats are optional
+
+            return HostStats(
+                host=host_name,
+                cpu_percent=data.get("cpu", 0),
+                mem_percent=data.get("mem", 0),
+                swap_percent=data.get("swap", 0),
+                load=data.get("load", 0),
+                disk_percent=disk_percent,
+                net_rx_rate=net_rx_rate,
+                net_tx_rate=net_tx_rate,
+            )
+    except httpx.TimeoutException:
+        return HostStats.from_error(host_name, "timeout")
+    except httpx.HTTPError as e:
+        return HostStats.from_error(host_name, str(e))
+    except Exception as e:
+        return HostStats.from_error(host_name, str(e))
+
+
+async def fetch_all_host_stats(
+    config: Config,
+    port: int = DEFAULT_GLANCES_PORT,
+) -> dict[str, HostStats]:
+    """Fetch stats from all hosts in parallel."""
+    tasks = [fetch_host_stats(name, host.address, port) for name, host in config.hosts.items()]
+    results = await asyncio.gather(*tasks)
+    return {stats.host: stats for stats in results}
+
+
+@dataclass
+class ContainerStats:
+    """Container statistics from Glances."""
+
+    name: str
+    host: str
+    status: str
+    image: str
+    cpu_percent: float
+    memory_usage: int  # bytes
+    memory_limit: int  # bytes
+    memory_percent: float
+    network_rx: int  # cumulative bytes received
+    network_tx: int  # cumulative bytes sent
+    uptime: str
+    ports: str
+    engine: str  # docker, podman, etc.
+    stack: str = ""  # compose project name (from docker labels)
+    service: str = ""  # compose service name (from docker labels)
+
+
+def _parse_container(data: dict[str, Any], host_name: str) -> ContainerStats:
+    """Parse container data from Glances API response."""
+    # Image can be a list or string
+    image = data.get("image", ["unknown"])
+    if isinstance(image, list):
+        image = image[0] if image else "unknown"
+
+    # Calculate memory percent
+    mem_usage = data.get("memory_usage", 0) or 0
+    mem_limit = data.get("memory_limit", 1) or 1  # Avoid division by zero
+    mem_percent = (mem_usage / mem_limit) * 100 if mem_limit > 0 else 0
+
+    # Network stats
+    network = data.get("network", {}) or {}
+    network_rx = network.get("cumulative_rx", 0) or 0
+    network_tx = network.get("cumulative_tx", 0) or 0
+
+    return ContainerStats(
+        name=data.get("name", "unknown"),
+        host=host_name,
+        status=data.get("status", "unknown"),
+        image=image,
+        cpu_percent=data.get("cpu_percent", 0) or 0,
+        memory_usage=mem_usage,
+        memory_limit=mem_limit,
+        memory_percent=mem_percent,
+        network_rx=network_rx,
+        network_tx=network_tx,
+        uptime=data.get("uptime", ""),
+        ports=data.get("ports", "") or "",
+        engine=data.get("engine", "docker"),
+    )
+
+
+async def fetch_container_stats(
+    host_name: str,
+    host_address: str,
+    port: int = DEFAULT_GLANCES_PORT,
+    request_timeout: float = 10.0,
+) -> tuple[list[ContainerStats] | None, str | None]:
+    """Fetch container stats from a single host's Glances API.
+
+    Returns:
+        (containers, error_message)
+        - Success: ([...], None)
+        - Failure: (None, "error message")
+
+    """
+    import httpx  # noqa: PLC0415
+
+    url = f"http://{host_address}:{port}/api/4/containers"
+
+    try:
+        async with httpx.AsyncClient(timeout=request_timeout) as client:
+            response = await client.get(url)
+            if not response.is_success:
+                return None, f"HTTP {response.status_code}: {response.reason_phrase}"
+            data = response.json()
+            return [_parse_container(c, host_name) for c in data], None
+    except httpx.ConnectError:
+        return None, "Connection refused (Glances offline?)"
+    except httpx.TimeoutException:
+        return None, "Connection timed out"
+    except Exception as e:
+        return None, str(e)
+
+
+async def fetch_all_container_stats(
+    config: Config,
+    port: int = DEFAULT_GLANCES_PORT,
+) -> list[ContainerStats]:
+    """Fetch container stats from all hosts in parallel, enriched with compose labels."""
+    from .executor import get_container_compose_labels  # noqa: PLC0415
+
+    async def fetch_host_data(
+        host_name: str,
+        host_address: str,
+    ) -> list[ContainerStats]:
+        # Fetch Glances stats and compose labels in parallel
+        stats_task = fetch_container_stats(host_name, host_address, port)
+        labels_task = get_container_compose_labels(config, host_name)
+        (containers, _), labels = await asyncio.gather(stats_task, labels_task)
+
+        if containers is None:
+            # Skip failed hosts in aggregate view
+            return []
+
+        # Enrich containers with compose labels (mutate in place)
+        for c in containers:
+            c.stack, c.service = labels.get(c.name, ("", ""))
+        return containers
+
+    tasks = [fetch_host_data(name, host.address) for name, host in config.hosts.items()]
+    results = await asyncio.gather(*tasks)
+    # Flatten list of lists
+    return [container for host_containers in results for container in host_containers]

src/compose_farm/logs.py

@@ -6,21 +6,22 @@ import json
 import tomllib
 from dataclasses import dataclass
 from datetime import UTC, datetime
-from typing import TYPE_CHECKING, Any
+from typing import TYPE_CHECKING
 
-from .executor import run_compose
+from .executor import run_command
 from .paths import xdg_config_home
 
 if TYPE_CHECKING:
-    from collections.abc import Awaitable, Callable, Iterable
+    from collections.abc import Iterable
     from pathlib import Path
 
     from .config import Config
-    from .executor import CommandResult
+
+# Separator used to split output sections
+_SECTION_SEPARATOR = "---CF-SEP---"
 
 
 DEFAULT_LOG_PATH = xdg_config_home() / "compose-farm" / "dockerfarm-log.toml"
-_DIGEST_HEX_LENGTH = 64
 
 
 @dataclass(frozen=True)
@@ -56,87 +57,97 @@ def _escape(value: str) -> str:
     return value.replace("\\", "\\\\").replace('"', '\\"')
 
 
-def _parse_images_output(raw: str) -> list[dict[str, Any]]:
-    """Parse `docker compose images --format json` output.
-
-    Handles both a JSON array and newline-separated JSON objects for robustness.
-    """
-    raw = raw.strip()
-    if not raw:
-        return []
-
+def _parse_image_digests(image_json: str) -> dict[str, str]:
+    """Parse docker image inspect JSON to build image tag -> digest map."""
+    if not image_json:
+        return {}
     try:
-        parsed = json.loads(raw)
+        image_data = json.loads(image_json)
     except json.JSONDecodeError:
-        objects = []
-        for line in raw.splitlines():
-            if not line.strip():
-                continue
-            objects.append(json.loads(line))
-        return objects
+        return {}
 
-    if isinstance(parsed, list):
-        return parsed
-    if isinstance(parsed, dict):
-        return [parsed]
-    return []
+    image_digests: dict[str, str] = {}
+    for img in image_data:
+        tags = img.get("RepoTags") or []
+        digests = img.get("RepoDigests") or []
+        digest = digests[0].split("@")[-1] if digests else img.get("Id", "")
+        for tag in tags:
+            image_digests[tag] = digest
+        if img.get("Id"):
+            image_digests[img["Id"]] = digest
+    return image_digests
 
 
-def _extract_image_fields(record: dict[str, Any]) -> tuple[str, str]:
-    """Extract image name and digest with fallbacks."""
-    image = record.get("Image") or record.get("Repository") or record.get("Name") or ""
-    tag = record.get("Tag") or record.get("Version")
-    if tag and ":" not in image.rsplit("/", 1)[-1]:
-        image = f"{image}:{tag}"
-
-    digest = (
-        record.get("Digest")
-        or record.get("Image ID")
-        or record.get("ImageID")
-        or record.get("ID")
-        or ""
-    )
-
-    if digest and not digest.startswith("sha256:") and len(digest) == _DIGEST_HEX_LENGTH:
-        digest = f"sha256:{digest}"
-
-    return image, digest
-
-
-async def collect_stack_entries(
+async def collect_stacks_entries_on_host(
     config: Config,
-    stack: str,
+    host_name: str,
+    stacks: set[str],
     *,
     now: datetime,
-    run_compose_fn: Callable[..., Awaitable[CommandResult]] = run_compose,
 ) -> list[SnapshotEntry]:
-    """Run `docker compose images` for a stack and normalize results."""
-    result = await run_compose_fn(config, stack, "images --format json", stream=False)
+    """Collect image entries for stacks on one host using 2 docker commands.
+
+    Uses `docker ps` to get running containers + their compose project labels,
+    then `docker image inspect` to get digests for all unique images.
+    Much faster than running N `docker compose images` commands.
+    """
+    if not stacks:
+        return []
+
+    host = config.hosts[host_name]
+
+    # Single SSH call with 2 docker commands:
+    # 1. Get project|image pairs from running containers
+    # 2. Get image info (including digests) for all unique images
+    command = (
+        f"docker ps --format '{{{{.Label \"com.docker.compose.project\"}}}}|{{{{.Image}}}}' && "
+        f"echo '{_SECTION_SEPARATOR}' && "
+        "docker image inspect $(docker ps --format '{{.Image}}' | sort -u) 2>/dev/null || true"
+    )
+    result = await run_command(host, command, host_name, stream=False, prefix="")
+
     if not result.success:
-        msg = result.stderr or f"compose images exited with {result.exit_code}"
-        error = f"[{stack}] Unable to read images: {msg}"
-        raise RuntimeError(error)
+        return []
 
-    records = _parse_images_output(result.stdout)
-    # Use first host for snapshots (multi-host stacks use same images on all hosts)
-    host_name = config.get_hosts(stack)[0]
-    compose_path = config.get_compose_path(stack)
+    # Split output into two sections
+    parts = result.stdout.split(_SECTION_SEPARATOR)
+    if len(parts) != 2:  # noqa: PLR2004
+        return []
 
-    entries: list[SnapshotEntry] = []
-    for record in records:
-        image, digest = _extract_image_fields(record)
-        if not digest:
+    container_lines, image_json = parts[0].strip(), parts[1].strip()
+
+    # Parse project|image pairs, filtering to only stacks we care about
+    stack_images: dict[str, set[str]] = {}
+    for line in container_lines.splitlines():
+        if "|" not in line:
             continue
-        entries.append(
-            SnapshotEntry(
-                stack=stack,
-                host=host_name,
-                compose_file=compose_path,
-                image=image,
-                digest=digest,
-                captured_at=now,
-            )
-        )
+        project, image = line.split("|", 1)
+        if project in stacks:
+            stack_images.setdefault(project, set()).add(image)
+
+    if not stack_images:
+        return []
+
+    # Parse image inspect JSON to build image -> digest map
+    image_digests = _parse_image_digests(image_json)
+
+    # Build entries
+    entries: list[SnapshotEntry] = []
+    for stack, images in stack_images.items():
+        for image in images:
+            digest = image_digests.get(image, "")
+            if digest:
+                entries.append(
+                    SnapshotEntry(
+                        stack=stack,
+                        host=host_name,
+                        compose_file=config.get_compose_path(stack),
+                        image=image,
+                        digest=digest,
+                        captured_at=now,
+                    )
+                )
+
     return entries
 
 

src/compose_farm/operations.py

@@ -76,29 +76,37 @@ def get_stack_paths(cfg: Config, stack: str) -> list[str]:
     return paths
 
 
-async def discover_stack_host(cfg: Config, stack: str) -> tuple[str, str | list[str] | None]:
-    """Discover where a stack is running.
+class StackDiscoveryResult(NamedTuple):
+    """Result of discovering where a stack is running across all hosts."""
 
-    For multi-host stacks, checks all assigned hosts in parallel.
-    For single-host, checks assigned host first, then others.
+    stack: str
+    configured_hosts: list[str]  # From config (where it SHOULD run)
+    running_hosts: list[str]  # From reality (where it IS running)
 
-    Returns (stack_name, host_or_hosts_or_none).
-    """
-    assigned_hosts = cfg.get_hosts(stack)
+    @property
+    def is_multi_host(self) -> bool:
+        """Check if this is a multi-host stack."""
+        return len(self.configured_hosts) > 1
 
-    if cfg.is_multi_host(stack):
-        # Check all assigned hosts in parallel
-        checks = await asyncio.gather(*[check_stack_running(cfg, stack, h) for h in assigned_hosts])
-        running = [h for h, is_running in zip(assigned_hosts, checks, strict=True) if is_running]
-        return stack, running if running else None
+    @property
+    def stray_hosts(self) -> list[str]:
+        """Hosts where stack is running but shouldn't be."""
+        return [h for h in self.running_hosts if h not in self.configured_hosts]
 
-    # Single-host: check assigned host first, then others
-    if await check_stack_running(cfg, stack, assigned_hosts[0]):
-        return stack, assigned_hosts[0]
-    for host in cfg.hosts:
-        if host != assigned_hosts[0] and await check_stack_running(cfg, stack, host):
-            return stack, host
-    return stack, None
+    @property
+    def missing_hosts(self) -> list[str]:
+        """Hosts where stack should be running but isn't."""
+        return [h for h in self.configured_hosts if h not in self.running_hosts]
+
+    @property
+    def is_stray(self) -> bool:
+        """Stack is running on unauthorized host(s)."""
+        return len(self.stray_hosts) > 0
+
+    @property
+    def is_duplicate(self) -> bool:
+        """Single-host stack running on multiple hosts."""
+        return not self.is_multi_host and len(self.running_hosts) > 1
 
 
 async def check_stack_requirements(
@@ -359,26 +367,33 @@ async def check_host_compatibility(
     return results
 
 
-async def stop_orphaned_stacks(cfg: Config) -> list[CommandResult]:
-    """Stop orphaned stacks (in state but not in config).
+async def _stop_stacks_on_hosts(
+    cfg: Config,
+    stacks_to_hosts: dict[str, list[str]],
+    label: str = "",
+) -> list[CommandResult]:
+    """Stop stacks on specific hosts.
 
-    Runs docker compose down on each stack on its tracked host(s).
-    Only removes from state on successful stop.
+    Shared helper for stop_orphaned_stacks and stop_stray_stacks.
+
+    Args:
+        cfg: Config object.
+        stacks_to_hosts: Dict mapping stack name to list of hosts to stop on.
+        label: Optional label for success message (e.g., "stray", "orphaned").
+
+    Returns:
+        List of CommandResults for each stack@host.
 
-    Returns list of CommandResults for each stack@host.
     """
-    orphaned = get_orphaned_stacks(cfg)
-    if not orphaned:
+    if not stacks_to_hosts:
         return []
 
     results: list[CommandResult] = []
     tasks: list[tuple[str, str, asyncio.Task[CommandResult]]] = []
+    suffix = f" ({label})" if label else ""
 
-    # Build list of (stack, host, task) for all orphaned stacks
-    for stack, hosts in orphaned.items():
-        host_list = hosts if isinstance(hosts, list) else [hosts]
-        for host in host_list:
-            # Skip hosts no longer in config
+    for stack, hosts in stacks_to_hosts.items():
+        for host in hosts:
             if host not in cfg.hosts:
                 print_warning(f"{stack}@{host}: host no longer in config, skipping")
                 results.append(
@@ -393,30 +408,48 @@ async def stop_orphaned_stacks(cfg: Config) -> list[CommandResult]:
             coro = run_compose_on_host(cfg, stack, host, "down")
             tasks.append((stack, host, asyncio.create_task(coro)))
 
-    # Run all down commands in parallel
-    if tasks:
-        for stack, host, task in tasks:
-            try:
-                result = await task
-                results.append(result)
-                if result.success:
-                    print_success(f"{stack}@{host}: stopped")
-                else:
-                    print_error(f"{stack}@{host}: {result.stderr or 'failed'}")
-            except Exception as e:
-                print_error(f"{stack}@{host}: {e}")
-                results.append(
-                    CommandResult(
-                        stack=f"{stack}@{host}",
-                        exit_code=1,
-                        success=False,
-                        stderr=str(e),
-                    )
+    for stack, host, task in tasks:
+        try:
+            result = await task
+            results.append(result)
+            if result.success:
+                print_success(f"{stack}@{host}: stopped{suffix}")
+            else:
+                print_error(f"{stack}@{host}: {result.stderr or 'failed'}")
+        except Exception as e:
+            print_error(f"{stack}@{host}: {e}")
+            results.append(
+                CommandResult(
+                    stack=f"{stack}@{host}",
+                    exit_code=1,
+                    success=False,
+                    stderr=str(e),
                 )
+            )
+
+    return results
+
+
+async def stop_orphaned_stacks(cfg: Config) -> list[CommandResult]:
+    """Stop orphaned stacks (in state but not in config).
+
+    Runs docker compose down on each stack on its tracked host(s).
+    Only removes from state on successful stop.
+
+    Returns list of CommandResults for each stack@host.
+    """
+    orphaned = get_orphaned_stacks(cfg)
+    if not orphaned:
+        return []
+
+    normalized: dict[str, list[str]] = {
+        stack: (hosts if isinstance(hosts, list) else [hosts]) for stack, hosts in orphaned.items()
+    }
+
+    results = await _stop_stacks_on_hosts(cfg, normalized)
 
     # Remove from state only for stacks where ALL hosts succeeded
-    for stack, hosts in orphaned.items():
-        host_list = hosts if isinstance(hosts, list) else [hosts]
+    for stack in normalized:
         all_succeeded = all(
             r.success for r in results if r.stack.startswith(f"{stack}@") or r.stack == stack
         )
@@ -424,3 +457,77 @@ async def stop_orphaned_stacks(cfg: Config) -> list[CommandResult]:
             remove_stack(cfg, stack)
 
     return results
+
+
+async def stop_stray_stacks(
+    cfg: Config,
+    strays: dict[str, list[str]],
+) -> list[CommandResult]:
+    """Stop stacks running on unauthorized hosts.
+
+    Args:
+        cfg: Config object.
+        strays: Dict mapping stack name to list of stray hosts.
+
+    Returns:
+        List of CommandResults for each stack@host stopped.
+
+    """
+    return await _stop_stacks_on_hosts(cfg, strays, label="stray")
+
+
+def build_discovery_results(
+    cfg: Config,
+    running_on_host: dict[str, set[str]],
+    stacks: list[str] | None = None,
+) -> tuple[dict[str, str | list[str]], dict[str, list[str]], dict[str, list[str]]]:
+    """Build discovery results from per-host running stacks.
+
+    Takes the raw data of which stacks are running on which hosts and
+    categorizes them into discovered (running correctly), strays (wrong host),
+    and duplicates (single-host stack on multiple hosts).
+
+    Args:
+        cfg: Config object.
+        running_on_host: Dict mapping host -> set of running stack names.
+        stacks: Optional list of stacks to check. Defaults to all configured stacks.
+
+    Returns:
+        Tuple of (discovered, strays, duplicates):
+        - discovered: stack -> host(s) where running correctly
+        - strays: stack -> list of unauthorized hosts
+        - duplicates: stack -> list of all hosts (for single-host stacks on multiple)
+
+    """
+    stack_list = stacks if stacks is not None else list(cfg.stacks)
+    all_hosts = list(running_on_host.keys())
+
+    # Build StackDiscoveryResult for each stack
+    results: list[StackDiscoveryResult] = [
+        StackDiscoveryResult(
+            stack=stack,
+            configured_hosts=cfg.get_hosts(stack),
+            running_hosts=[h for h in all_hosts if stack in running_on_host[h]],
+        )
+        for stack in stack_list
+    ]
+
+    discovered: dict[str, str | list[str]] = {}
+    strays: dict[str, list[str]] = {}
+    duplicates: dict[str, list[str]] = {}
+
+    for result in results:
+        correct_hosts = [h for h in result.running_hosts if h in result.configured_hosts]
+        if correct_hosts:
+            if result.is_multi_host:
+                discovered[result.stack] = correct_hosts
+            else:
+                discovered[result.stack] = correct_hosts[0]
+
+        if result.is_stray:
+            strays[result.stack] = result.stray_hosts
+
+        if result.is_duplicate:
+            duplicates[result.stack] = result.running_hosts
+
+    return discovered, strays, duplicates

src/compose_farm/paths.py

@@ -11,9 +11,19 @@ def xdg_config_home() -> Path:
     return Path(os.environ.get("XDG_CONFIG_HOME", Path.home() / ".config"))
 
 
+def config_dir() -> Path:
+    """Get the compose-farm config directory."""
+    return xdg_config_home() / "compose-farm"
+
+
 def default_config_path() -> Path:
     """Get the default user config path."""
-    return xdg_config_home() / "compose-farm" / "compose-farm.yaml"
+    return config_dir() / "compose-farm.yaml"
+
+
+def backup_dir() -> Path:
+    """Get the backup directory for file edits."""
+    return config_dir() / "backups"
 
 
 def config_search_paths() -> list[Path]:

src/compose_farm/registry.py

@@ -0,0 +1,220 @@
+"""Container registry API client for tag discovery."""
+
+from __future__ import annotations
+
+import re
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING
+
+if TYPE_CHECKING:
+    import httpx
+
+# Image reference pattern: [registry/][namespace/]name[:tag][@digest]
+IMAGE_PATTERN = re.compile(
+    r"^(?:(?P<registry>[^/]+\.[^/]+)/)?(?:(?P<namespace>[^/:@]+)/)?(?P<name>[^/:@]+)(?::(?P<tag>[^@]+))?(?:@(?P<digest>.+))?$"
+)
+
+# Docker Hub aliases
+DOCKER_HUB_ALIASES = frozenset(
+    {"docker.io", "index.docker.io", "registry.hub.docker.com", "registry-1.docker.io"}
+)
+
+# Token endpoints per registry: (url, extra_params)
+TOKEN_ENDPOINTS: dict[str, tuple[str, dict[str, str]]] = {
+    "docker.io": ("https://auth.docker.io/token", {"service": "registry.docker.io"}),
+    "ghcr.io": ("https://ghcr.io/token", {}),
+}
+
+# Registry URL overrides (Docker Hub uses a different host for API)
+REGISTRY_URLS: dict[str, str] = {
+    "docker.io": "https://registry-1.docker.io",
+}
+
+HTTP_OK = 200
+
+MANIFEST_ACCEPT = (
+    "application/vnd.docker.distribution.manifest.v2+json, "
+    "application/vnd.oci.image.manifest.v1+json, "
+    "application/vnd.oci.image.index.v1+json"
+)
+
+
+@dataclass(frozen=True)
+class ImageRef:
+    """Parsed container image reference."""
+
+    registry: str
+    namespace: str
+    name: str
+    tag: str
+    digest: str | None = None
+
+    @property
+    def full_name(self) -> str:
+        """Full image name with namespace."""
+        return f"{self.namespace}/{self.name}" if self.namespace else self.name
+
+    @property
+    def display_name(self) -> str:
+        """Display name (omits docker.io/library for official images)."""
+        if self.registry in DOCKER_HUB_ALIASES:
+            if self.namespace == "library":
+                return self.name
+            return self.full_name
+        return f"{self.registry}/{self.full_name}"
+
+    @classmethod
+    def parse(cls, image: str) -> ImageRef:
+        """Parse image string into components."""
+        match = IMAGE_PATTERN.match(image)
+        if not match:
+            return cls("docker.io", "library", image.split(":")[0].split("@")[0], "latest")
+
+        groups = match.groupdict()
+        registry = groups.get("registry") or "docker.io"
+        namespace = groups.get("namespace") or ""
+        name = groups.get("name") or image
+        tag = groups.get("tag") or "latest"
+        digest = groups.get("digest")
+
+        # Docker Hub official images have implicit "library" namespace
+        if registry in DOCKER_HUB_ALIASES and not namespace:
+            namespace = "library"
+
+        return cls(registry, namespace, name, tag, digest)
+
+
+@dataclass
+class TagCheckResult:
+    """Result of checking tags for an image."""
+
+    image: ImageRef
+    current_digest: str
+    available_updates: list[str] = field(default_factory=list)
+    error: str | None = None
+
+
+class RegistryClient:
+    """Unified OCI Distribution API client."""
+
+    def __init__(self, registry: str) -> None:
+        """Initialize for a specific registry."""
+        self.registry = registry.lower()
+        # Normalize Docker Hub aliases
+        if self.registry in DOCKER_HUB_ALIASES:
+            self.registry = "docker.io"
+
+        self.registry_url = REGISTRY_URLS.get(self.registry, f"https://{self.registry}")
+        self._token_cache: dict[str, str] = {}
+
+    async def _get_token(self, image: ImageRef, client: httpx.AsyncClient) -> str | None:
+        """Get auth token for the registry (cached per image)."""
+        cache_key = image.full_name
+        if cache_key in self._token_cache:
+            return self._token_cache[cache_key]
+
+        endpoint = TOKEN_ENDPOINTS.get(self.registry)
+        if not endpoint:
+            return None  # No auth needed or unknown registry
+
+        url, extra_params = endpoint
+        params = {"scope": f"repository:{image.full_name}:pull", **extra_params}
+        resp = await client.get(url, params=params)
+
+        if resp.status_code == HTTP_OK:
+            token: str | None = resp.json().get("token")
+            if token:
+                self._token_cache[cache_key] = token
+            return token
+        return None
+
+    async def get_tags(self, image: ImageRef, client: httpx.AsyncClient) -> list[str]:
+        """Fetch available tags for an image."""
+        headers = {}
+        token = await self._get_token(image, client)
+        if token:
+            headers["Authorization"] = f"Bearer {token}"
+
+        url = f"{self.registry_url}/v2/{image.full_name}/tags/list"
+        resp = await client.get(url, headers=headers)
+
+        if resp.status_code != HTTP_OK:
+            return []
+        tags: list[str] = resp.json().get("tags", [])
+        return tags
+
+    async def get_digest(self, image: ImageRef, tag: str, client: httpx.AsyncClient) -> str | None:
+        """Get digest for a specific tag."""
+        headers = {"Accept": MANIFEST_ACCEPT}
+        token = await self._get_token(image, client)
+        if token:
+            headers["Authorization"] = f"Bearer {token}"
+
+        url = f"{self.registry_url}/v2/{image.full_name}/manifests/{tag}"
+        resp = await client.head(url, headers=headers)
+
+        if resp.status_code == HTTP_OK:
+            digest: str | None = resp.headers.get("docker-content-digest")
+            return digest
+        return None
+
+
+def _parse_version(tag: str) -> tuple[int, ...] | None:
+    """Parse version string into comparable tuple."""
+    tag = tag.lstrip("vV")
+    parts = tag.split(".")
+    try:
+        return tuple(int(p) for p in parts)
+    except ValueError:
+        return None
+
+
+def _find_updates(current_tag: str, tags: list[str]) -> list[str]:
+    """Find tags newer than current based on version comparison."""
+    current_version = _parse_version(current_tag)
+    if current_version is None:
+        return []
+
+    updates = []
+    for tag in tags:
+        tag_version = _parse_version(tag)
+        if tag_version and tag_version > current_version:
+            updates.append(tag)
+
+    updates.sort(key=lambda t: _parse_version(t) or (), reverse=True)
+    return updates
+
+
+async def check_image_updates(
+    image_str: str,
+    client: httpx.AsyncClient,
+) -> TagCheckResult:
+    """Check if newer versions are available for an image.
+
+    Args:
+        image_str: Image string like "nginx:1.25" or "ghcr.io/user/repo:tag"
+        client: httpx async client
+
+    Returns:
+        TagCheckResult with available updates
+
+    """
+    image = ImageRef.parse(image_str)
+    registry_client = RegistryClient(image.registry)
+
+    try:
+        tags = await registry_client.get_tags(image, client)
+        updates = _find_updates(image.tag, tags)
+        current_digest = await registry_client.get_digest(image, image.tag, client) or ""
+
+        return TagCheckResult(
+            image=image,
+            current_digest=current_digest,
+            available_updates=updates,
+        )
+    except Exception as e:
+        return TagCheckResult(
+            image=image,
+            current_digest="",
+            error=str(e),
+        )

src/compose_farm/web/app.py

@@ -6,15 +6,16 @@ import asyncio
 import logging
 import sys
 from contextlib import asynccontextmanager, suppress
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any, cast
 
 from fastapi import FastAPI
+from fastapi.middleware.gzip import GZipMiddleware
 from fastapi.staticfiles import StaticFiles
 from pydantic import ValidationError
 from rich.logging import RichHandler
 
 from compose_farm.web.deps import STATIC_DIR, get_config
-from compose_farm.web.routes import actions, api, pages
+from compose_farm.web.routes import actions, api, containers, pages
 from compose_farm.web.streaming import TASK_TTL_SECONDS, cleanup_stale_tasks
 
 # Configure logging with Rich handler for compose_farm.web modules
@@ -64,10 +65,14 @@ def create_app() -> FastAPI:
         lifespan=lifespan,
     )
 
+    # Enable Gzip compression for faster transfers over slow networks
+    app.add_middleware(cast("Any", GZipMiddleware), minimum_size=1000)
+
     # Mount static files
     app.mount("/static", StaticFiles(directory=str(STATIC_DIR)), name="static")
 
     app.include_router(pages.router)
+    app.include_router(containers.router)
     app.include_router(api.router, prefix="/api")
     app.include_router(actions.router, prefix="/api")
 

src/compose_farm/web/cdn.py

@@ -39,6 +39,14 @@ CDN_ASSETS: dict[str, tuple[str, str]] = {
         "xterm-fit.js",
         "application/javascript",
     ),
+    "https://unpkg.com/idiomorph/dist/idiomorph.min.js": (
+        "idiomorph.js",
+        "application/javascript",
+    ),
+    "https://unpkg.com/idiomorph/dist/idiomorph-ext.min.js": (
+        "idiomorph-ext.js",
+        "application/javascript",
+    ),
     # Monaco editor - dynamically loaded by app.js
     "https://cdn.jsdelivr.net/npm/monaco-editor@0.52.2/min/vs/loader.js": (
         "monaco-loader.js",

src/compose_farm/web/deps.py

@@ -38,7 +38,17 @@ def get_templates() -> Jinja2Templates:
 def extract_config_error(exc: Exception) -> str:
     """Extract a user-friendly error message from a config exception."""
     if isinstance(exc, ValidationError):
-        return "; ".join(err.get("msg", str(err)) for err in exc.errors())
+        parts = []
+        for err in exc.errors():
+            msg = err.get("msg", str(err))
+            loc = err.get("loc", ())
+            if loc:
+                # Format location as dot-separated path (e.g., "hosts.nas.port")
+                loc_str = ".".join(str(part) for part in loc)
+                parts.append(f"{loc_str}: {msg}")
+            else:
+                parts.append(msg)
+        return "; ".join(parts)
     return str(exc)
 
 

src/compose_farm/web/routes/__init__.py

@@ -1,5 +1,5 @@
 """Web routes."""
 
-from compose_farm.web.routes import actions, api, pages
+from compose_farm.web.routes import actions, api, containers, pages
 
-__all__ = ["actions", "api", "pages"]
+__all__ = ["actions", "api", "containers", "pages"]

src/compose_farm/web/routes/api.py

@@ -21,7 +21,8 @@ from fastapi.responses import HTMLResponse
 
 from compose_farm.compose import get_container_name
 from compose_farm.executor import is_local, run_compose_on_host, ssh_connect_kwargs
-from compose_farm.paths import find_config_path
+from compose_farm.glances import fetch_all_host_stats
+from compose_farm.paths import backup_dir, find_config_path
 from compose_farm.state import load_state
 from compose_farm.web.deps import get_config, get_templates
 
@@ -41,26 +42,30 @@ def _validate_yaml(content: str) -> None:
 def _backup_file(file_path: Path) -> Path | None:
     """Create a timestamped backup of a file if it exists and content differs.
 
-    Backups are stored in a .backups directory alongside the file.
+    Backups are stored in XDG config dir under compose-farm/backups/.
+    The original file's absolute path is mirrored in the backup directory.
     Returns the backup path if created, None if no backup was needed.
     """
     if not file_path.exists():
         return None
 
-    # Create backup directory
-    backup_dir = file_path.parent / ".backups"
-    backup_dir.mkdir(exist_ok=True)
+    # Create backup directory mirroring original path structure
+    # e.g., /opt/stacks/plex/compose.yaml -> ~/.config/compose-farm/backups/opt/stacks/plex/
+    # On Windows: C:\Users\foo\stacks -> backups/Users/foo/stacks
+    resolved = file_path.resolve()
+    file_backup_dir = backup_dir() / resolved.parent.relative_to(resolved.anchor)
+    file_backup_dir.mkdir(parents=True, exist_ok=True)
 
     # Generate timestamped backup filename
     timestamp = datetime.now(tz=UTC).strftime("%Y%m%d_%H%M%S")
     backup_name = f"{file_path.name}.{timestamp}"
-    backup_path = backup_dir / backup_name
+    backup_path = file_backup_dir / backup_name
 
     # Copy current content to backup
     backup_path.write_text(file_path.read_text())
 
     # Clean up old backups (keep last 200)
-    backups = sorted(backup_dir.glob(f"{file_path.name}.*"), reverse=True)
+    backups = sorted(file_backup_dir.glob(f"{file_path.name}.*"), reverse=True)
     for old_backup in backups[200:]:
         old_backup.unlink()
 
@@ -381,3 +386,19 @@ async def write_console_file(
     except Exception as e:
         logger.exception("Failed to write file %s to host %s", path, host)
         raise HTTPException(status_code=500, detail=str(e)) from e
+
+
+@router.get("/glances", response_class=HTMLResponse)
+async def get_glances_stats() -> HTMLResponse:
+    """Get resource stats from Glances for all hosts."""
+    config = get_config()
+
+    if not config.glances_stack:
+        return HTMLResponse("")  # Glances not configured
+
+    stats = await fetch_all_host_stats(config)
+
+    templates = get_templates()
+    template = templates.env.get_template("partials/glances.html")
+    html = template.render(stats=stats)
+    return HTMLResponse(html)

src/compose_farm/web/routes/containers.py

@@ -0,0 +1,370 @@
+"""Container dashboard routes using Glances API."""
+
+from __future__ import annotations
+
+import html
+import re
+from typing import TYPE_CHECKING
+from urllib.parse import quote
+
+import humanize
+from fastapi import APIRouter, Request
+from fastapi.responses import HTMLResponse, JSONResponse
+
+from compose_farm.executor import TTLCache
+from compose_farm.glances import ContainerStats, fetch_all_container_stats
+from compose_farm.registry import DOCKER_HUB_ALIASES, ImageRef
+from compose_farm.web.deps import get_config, get_templates
+
+router = APIRouter(tags=["containers"])
+
+if TYPE_CHECKING:
+    from compose_farm.registry import TagCheckResult
+
+# Cache registry update checks for 5 minutes (300 seconds)
+# Registry calls are slow and often rate-limited
+_update_check_cache = TTLCache(ttl_seconds=300.0)
+
+# Minimum parts needed to infer stack/service from container name
+MIN_NAME_PARTS = 2
+
+# HTML for "no update info" dash
+_DASH_HTML = '<span class="text-xs opacity-50">-</span>'
+
+
+def _format_bytes(bytes_val: int) -> str:
+    """Format bytes to human readable string."""
+    return humanize.naturalsize(bytes_val, binary=True, format="%.1f")
+
+
+def _parse_image(image: str) -> tuple[str, str]:
+    """Parse image string into (name, tag)."""
+    # Handle registry prefix (e.g., ghcr.io/user/repo:tag)
+    if ":" in image:
+        # Find last colon that's not part of port
+        parts = image.rsplit(":", 1)
+        if "/" in parts[-1]:
+            # The "tag" contains a slash, so it's probably a port
+            return image, "latest"
+        return parts[0], parts[1]
+    return image, "latest"
+
+
+def _infer_stack_service(name: str) -> tuple[str, str]:
+    """Fallback: infer stack and service from container name.
+
+    Used when compose labels are not available.
+    Docker Compose naming conventions:
+    - Default: {project}_{service}_{instance} or {project}-{service}-{instance}
+    - Custom: {container_name} from compose file
+    """
+    # Try underscore separator first (older compose)
+    if "_" in name:
+        parts = name.split("_")
+        if len(parts) >= MIN_NAME_PARTS:
+            return parts[0], parts[1]
+    # Try hyphen separator (newer compose)
+    if "-" in name:
+        parts = name.split("-")
+        if len(parts) >= MIN_NAME_PARTS:
+            return parts[0], "-".join(parts[1:-1]) if len(parts) > MIN_NAME_PARTS else parts[1]
+    # Fallback: use name as both stack and service
+    return name, name
+
+
+@router.get("/live-stats", response_class=HTMLResponse)
+async def containers_page(request: Request) -> HTMLResponse:
+    """Container dashboard page."""
+    config = get_config()
+    templates = get_templates()
+
+    # Check if Glances is configured
+    glances_enabled = config.glances_stack is not None
+
+    return templates.TemplateResponse(
+        "containers.html",
+        {
+            "request": request,
+            "glances_enabled": glances_enabled,
+            "hosts": sorted(config.hosts.keys()) if glances_enabled else [],
+        },
+    )
+
+
+_STATUS_CLASSES = {
+    "running": "badge badge-success badge-sm",
+    "exited": "badge badge-error badge-sm",
+    "paused": "badge badge-warning badge-sm",
+}
+
+
+def _status_class(status: str) -> str:
+    """Get CSS class for status badge."""
+    return _STATUS_CLASSES.get(status.lower(), "badge badge-ghost badge-sm")
+
+
+def _progress_class(percent: float) -> str:
+    """Get CSS class for progress bar color."""
+    if percent > 80:  # noqa: PLR2004
+        return "bg-error"
+    if percent > 50:  # noqa: PLR2004
+        return "bg-warning"
+    return "bg-success"
+
+
+def _render_update_cell(image: str, tag: str) -> str:
+    """Render update check cell with client-side batch updates."""
+    encoded_image = quote(image, safe="")
+    encoded_tag = quote(tag, safe="")
+    cached_html = _update_check_cache.get(f"{image}:{tag}")
+    inner = cached_html if cached_html is not None else _DASH_HTML
+    return (
+        f"""<td class="update-cell" data-image="{encoded_image}" data-tag="{encoded_tag}">"""
+        f"{inner}</td>"
+    )
+
+
+def _image_web_url(image: str) -> str | None:
+    """Return a human-friendly registry URL for an image (without tag)."""
+    ref = ImageRef.parse(image)
+    if ref.registry in DOCKER_HUB_ALIASES:
+        if ref.namespace == "library":
+            return f"https://hub.docker.com/_/{ref.name}"
+        return f"https://hub.docker.com/r/{ref.namespace}/{ref.name}"
+    return f"https://{ref.registry}/{ref.full_name}"
+
+
+def _render_row(c: ContainerStats, idx: int | str) -> str:
+    """Render a single container as an HTML table row."""
+    image_name, tag = _parse_image(c.image)
+    stack = c.stack if c.stack else _infer_stack_service(c.name)[0]
+    service = c.service if c.service else _infer_stack_service(c.name)[1]
+
+    cpu = c.cpu_percent
+    mem = c.memory_percent
+    cpu_class = _progress_class(cpu)
+    mem_class = _progress_class(mem)
+
+    # Highlight rows with high resource usage
+    high_cpu = cpu > 80  # noqa: PLR2004
+    high_mem = mem > 90  # noqa: PLR2004
+    row_class = "high-usage" if (high_cpu or high_mem) else ""
+
+    uptime_sec = _parse_uptime_seconds(c.uptime)
+    actions = _render_actions(stack)
+    update_cell = _render_update_cell(image_name, tag)
+    image_label = f"{image_name}:{tag}"
+    image_url = _image_web_url(image_name)
+    if image_url:
+        image_html = (
+            f'<a href="{image_url}" target="_blank" rel="noopener noreferrer" '
+            f'class="link link-hover">'
+            f'<code class="text-xs bg-base-200 px-1 rounded">{image_label}</code></a>'
+        )
+    else:
+        image_html = f'<code class="text-xs bg-base-200 px-1 rounded">{image_label}</code>'
+    # Render as single line to avoid whitespace nodes in DOM
+    row_id = f"c-{c.host}-{c.name}"
+    class_attr = f' class="{row_class}"' if row_class else ""
+    return (
+        f'<tr id="{row_id}" data-host="{c.host}"{class_attr}><td class="text-xs opacity-50">{idx}</td>'
+        f'<td data-sort="{stack.lower()}"><a href="/stack/{stack}" class="link link-hover link-primary" hx-boost="true">{stack}</a></td>'
+        f'<td data-sort="{service.lower()}" class="text-xs opacity-70">{service}</td>'
+        f"<td>{actions}</td>"
+        f'<td data-sort="{c.host.lower()}"><span class="badge badge-outline badge-xs">{c.host}</span></td>'
+        f'<td data-sort="{c.image.lower()}">{image_html}</td>'
+        f"{update_cell}"
+        f'<td data-sort="{c.status.lower()}"><span class="{_status_class(c.status)}">{c.status}</span></td>'
+        f'<td data-sort="{uptime_sec}" class="text-xs text-right font-mono">{c.uptime or "-"}</td>'
+        f'<td data-sort="{cpu}" class="text-right font-mono"><div class="flex flex-col items-end gap-0.5"><div class="w-12 h-2 bg-base-300 rounded-full overflow-hidden"><div class="h-full {cpu_class}" style="width: {min(cpu, 100)}%"></div></div><span class="text-xs">{cpu:.0f}%</span></div></td>'
+        f'<td data-sort="{c.memory_usage}" class="text-right font-mono"><div class="flex flex-col items-end gap-0.5"><div class="w-12 h-2 bg-base-300 rounded-full overflow-hidden"><div class="h-full {mem_class}" style="width: {min(mem, 100)}%"></div></div><span class="text-xs">{_format_bytes(c.memory_usage)}</span></div></td>'
+        f'<td data-sort="{c.network_rx + c.network_tx}" class="text-xs text-right font-mono">↓{_format_bytes(c.network_rx)} ↑{_format_bytes(c.network_tx)}</td>'
+        "</tr>"
+    )
+
+
+def _render_actions(stack: str) -> str:
+    """Render actions dropdown for a container row."""
+    return f"""<button class="btn btn-circle btn-ghost btn-xs" onclick="openActionMenu(event, '{stack}')" aria-label="Actions for {stack}">
+<svg class="h-4 w-4"><use href="#icon-menu" /></svg>
+</button>"""
+
+
+def _parse_uptime_seconds(uptime: str) -> int:
+    """Parse uptime string to seconds for sorting."""
+    if not uptime:
+        return 0
+    uptime = uptime.lower().strip()
+    # Handle "a/an" as 1
+    uptime = uptime.replace("an ", "1 ").replace("a ", "1 ")
+
+    total = 0
+    multipliers = {
+        "second": 1,
+        "minute": 60,
+        "hour": 3600,
+        "day": 86400,
+        "week": 604800,
+        "month": 2592000,
+        "year": 31536000,
+    }
+    for match in re.finditer(r"(\d+)\s*(\w+)", uptime):
+        num = int(match.group(1))
+        unit = match.group(2).rstrip("s")  # Remove plural 's'
+        total += num * multipliers.get(unit, 0)
+    return total
+
+
+@router.get("/api/containers/rows", response_class=HTMLResponse)
+async def get_containers_rows() -> HTMLResponse:
+    """Get container table rows as HTML for HTMX.
+
+    Each cell has data-sort attribute for instant client-side sorting.
+    """
+    config = get_config()
+
+    if not config.glances_stack:
+        return HTMLResponse(
+            '<tr><td colspan="12" class="text-center text-error">Glances not configured</td></tr>'
+        )
+
+    containers = await fetch_all_container_stats(config)
+
+    if not containers:
+        return HTMLResponse(
+            '<tr><td colspan="12" class="text-center py-4 opacity-60">No containers found</td></tr>'
+        )
+
+    rows = "\n".join(_render_row(c, i + 1) for i, c in enumerate(containers))
+    return HTMLResponse(rows)
+
+
+@router.get("/api/containers/rows/{host_name}", response_class=HTMLResponse)
+async def get_containers_rows_by_host(host_name: str) -> HTMLResponse:
+    """Get container rows for a specific host.
+
+    Returns immediately with Glances data. Stack/service are inferred from
+    container names for instant display (no SSH wait).
+    """
+    import logging  # noqa: PLC0415
+    import time  # noqa: PLC0415
+
+    from compose_farm.executor import get_container_compose_labels  # noqa: PLC0415
+    from compose_farm.glances import fetch_container_stats  # noqa: PLC0415
+
+    logger = logging.getLogger(__name__)
+    config = get_config()
+
+    if host_name not in config.hosts:
+        return HTMLResponse("")
+
+    host = config.hosts[host_name]
+
+    t0 = time.monotonic()
+    containers, error = await fetch_container_stats(host_name, host.address)
+    t1 = time.monotonic()
+    fetch_ms = (t1 - t0) * 1000
+
+    if containers is None:
+        logger.error(
+            "Failed to fetch stats for %s in %.1fms: %s",
+            host_name,
+            fetch_ms,
+            error,
+        )
+        return HTMLResponse(
+            f'<tr class="text-error"><td colspan="12" class="text-center py-2">Error: {error}</td></tr>'
+        )
+
+    if not containers:
+        return HTMLResponse("")  # No rows for this host
+
+    labels = await get_container_compose_labels(config, host_name)
+    for c in containers:
+        stack, service = labels.get(c.name, ("", ""))
+        if not stack or not service:
+            stack, service = _infer_stack_service(c.name)
+        c.stack, c.service = stack, service
+
+    # Only show containers from stacks in config (filters out orphaned/unknown stacks)
+    containers = [c for c in containers if not c.stack or c.stack in config.stacks]
+
+    # Use placeholder index (will be renumbered by JS after all hosts load)
+    rows = "\n".join(_render_row(c, "-") for c in containers)
+    t2 = time.monotonic()
+    render_ms = (t2 - t1) * 1000
+
+    logger.info(
+        "Loaded %d rows for %s in %.1fms (fetch) + %.1fms (render)",
+        len(containers),
+        host_name,
+        fetch_ms,
+        render_ms,
+    )
+    return HTMLResponse(rows)
+
+
+def _render_update_badge(result: TagCheckResult) -> str:
+    if result.error:
+        return _DASH_HTML
+    if result.available_updates:
+        updates = result.available_updates
+        count = len(updates)
+        title = f"Newer: {', '.join(updates[:3])}" + ("..." if count > 3 else "")  # noqa: PLR2004
+        tip = html.escape(title, quote=True)
+        return (
+            f'<span class="tooltip" data-tip="{tip}">'
+            f'<span class="badge badge-warning badge-xs cursor-help">{count} new</span>'
+            "</span>"
+        )
+    return '<span class="tooltip" data-tip="Up to date"><span class="text-success text-xs">✓</span></span>'
+
+
+@router.post("/api/containers/check-updates", response_class=JSONResponse)
+async def check_container_updates_batch(request: Request) -> JSONResponse:
+    """Batch update checks for a list of images.
+
+    Payload: {"items": [{"image": "...", "tag": "..."}, ...]}
+    Returns: {"results": [{"image": "...", "tag": "...", "html": "..."}, ...]}
+    """
+    import httpx  # noqa: PLC0415
+
+    payload = await request.json()
+    items = payload.get("items", []) if isinstance(payload, dict) else []
+    if not items:
+        return JSONResponse({"results": []})
+
+    results = []
+
+    from compose_farm.registry import check_image_updates  # noqa: PLC0415
+
+    async with httpx.AsyncClient(timeout=10.0) as client:
+        for item in items:
+            image = item.get("image", "")
+            tag = item.get("tag", "")
+            full_image = f"{image}:{tag}"
+            if not image or not tag:
+                results.append({"image": image, "tag": tag, "html": _DASH_HTML})
+                continue
+
+            # NOTE: Tag-based checks cannot detect digest changes for moving tags
+            # like "latest". A future improvement could compare remote vs local
+            # digests using dockerfarm-log.toml (from `cf refresh`) or a per-host
+            # digest lookup.
+
+            cached_html: str | None = _update_check_cache.get(full_image)
+            if cached_html is not None:
+                results.append({"image": image, "tag": tag, "html": cached_html})
+                continue
+
+            try:
+                result = await check_image_updates(full_image, client)
+                html = _render_update_badge(result)
+                _update_check_cache.set(full_image, html)
+            except Exception:
+                _update_check_cache.set(full_image, _DASH_HTML, ttl_seconds=60.0)
+                html = _DASH_HTML
+
+            results.append({"image": image, "tag": tag, "html": html})
+
+    return JSONResponse({"results": results})

src/compose_farm/web/routes/pages.py

@@ -91,8 +91,8 @@ async def index(request: Request) -> HTMLResponse:
     # Get state
     deployed = load_state(config)
 
-    # Stats
-    running_count = len(deployed)
+    # Stats (only count stacks that are both in config AND deployed)
+    running_count = sum(1 for stack in deployed if stack in config.stacks)
     stopped_count = len(config.stacks) - running_count
 
     # Pending operations
@@ -250,7 +250,8 @@ async def stats_partial(request: Request) -> HTMLResponse:
     templates = get_templates()
 
     deployed = load_state(config)
-    running_count = len(deployed)
+    # Only count stacks that are both in config AND deployed
+    running_count = sum(1 for stack in deployed if stack in config.stacks)
     stopped_count = len(config.stacks) - running_count
 
     return templates.TemplateResponse(

src/compose_farm/web/static/app.js

@@ -9,7 +9,6 @@
 // ANSI escape codes for terminal output
 const ANSI = {
     RED: '\x1b[31m',
-    GREEN: '\x1b[32m',
     DIM: '\x1b[2m',
     RESET: '\x1b[0m',
     CRLF: '\r\n'
@@ -122,7 +121,6 @@ function whenXtermReady(callback, maxAttempts = 20) {
     };
     tryInit(maxAttempts);
 }
-window.whenXtermReady = whenXtermReady;
 
 // ============================================================================
 // TERMINAL
@@ -209,8 +207,6 @@ function initTerminal(elementId, taskId) {
     return { term, ws };
 }
 
-window.initTerminal = initTerminal;
-
 /**
  * Initialize an interactive exec terminal
  */
@@ -223,7 +219,9 @@ function initExecTerminal(stack, container, host) {
         return;
     }
 
+    // Unhide the terminal container first, then expand/scroll
     containerEl.classList.remove('hidden');
+    expandCollapse(document.getElementById('exec-collapse'), containerEl);
 
     // Clean up existing (use wrapper's dispose to clean up ResizeObserver)
     if (execWs) { execWs.close(); execWs = null; }
@@ -259,17 +257,42 @@ function initExecTerminal(stack, container, host) {
 
 window.initExecTerminal = initExecTerminal;
 
+/**
+ * Expand a collapse component and scroll to a target element
+ * @param {HTMLInputElement} toggle - The checkbox input that controls the collapse
+ * @param {HTMLElement} [scrollTarget] - Element to scroll to (defaults to collapse container)
+ */
+function expandCollapse(toggle, scrollTarget = null) {
+    if (!toggle) return;
+
+    // Find the parent collapse container
+    const collapse = toggle.closest('.collapse');
+    if (!collapse) return;
+
+    const target = scrollTarget || collapse;
+    const scrollToTarget = () => {
+        target.scrollIntoView({ behavior: 'smooth', block: 'start' });
+    };
+
+    if (!toggle.checked) {
+        // Collapsed - expand first, then scroll after transition
+        const onTransitionEnd = () => {
+            collapse.removeEventListener('transitionend', onTransitionEnd);
+            scrollToTarget();
+        };
+        collapse.addEventListener('transitionend', onTransitionEnd);
+        toggle.checked = true;
+    } else {
+        // Already expanded - just scroll
+        scrollToTarget();
+    }
+}
+
 /**
  * Expand terminal collapse and scroll to it
  */
 function expandTerminal() {
-    const toggle = document.getElementById('terminal-toggle');
-    if (toggle) toggle.checked = true;
-
-    const collapse = document.getElementById('terminal-collapse');
-    if (collapse) {
-        collapse.scrollIntoView({ behavior: 'smooth', block: 'start' });
-    }
+    expandCollapse(document.getElementById('terminal-toggle'));
 }
 
 /**
@@ -405,7 +428,7 @@ function initMonacoEditors() {
  * Save all editors
  */
 async function saveAllEditors() {
-    const saveBtn = document.getElementById('save-btn') || document.getElementById('save-config-btn');
+    const saveBtn = getSaveButton();
     const results = [];
 
     for (const [id, editor] of Object.entries(editors)) {
@@ -441,12 +464,16 @@ async function saveAllEditors() {
  * Initialize save button handler
  */
 function initSaveButton() {
-    const saveBtn = document.getElementById('save-btn') || document.getElementById('save-config-btn');
+    const saveBtn = getSaveButton();
     if (!saveBtn) return;
 
     saveBtn.onclick = saveAllEditors;
 }
 
+function getSaveButton() {
+    return document.getElementById('save-btn') || document.getElementById('save-config-btn');
+}
+
 // ============================================================================
 // UI HELPERS
 // ============================================================================
@@ -531,6 +558,7 @@ function playFabIntro() {
         }
         htmx.ajax('GET', url, {target: '#main-content', select: '#main-content', swap: 'outerHTML'}).then(() => {
             history.pushState({}, '', url);
+            window.scrollTo(0, 0);
             afterNav?.();
         });
     };
@@ -539,6 +567,7 @@ function playFabIntro() {
         if (window.location.pathname !== '/') {
             await htmx.ajax('GET', '/', {target: '#main-content', select: '#main-content', swap: 'outerHTML'});
             history.pushState({}, '', '/');
+            window.scrollTo(0, 0);
         }
         htmx.ajax('POST', `/api/${endpoint}`, {swap: 'none'});
     };
@@ -578,9 +607,11 @@ function playFabIntro() {
             cmd('action', 'Update All', 'Update all stacks', dashboardAction('update-all'), icons.refresh_cw),
             cmd('app', 'Theme', 'Change color theme', openThemePicker, icons.palette),
             cmd('app', 'Dashboard', 'Go to dashboard', nav('/'), icons.home),
+            cmd('app', 'Live Stats', 'View all containers across hosts', nav('/live-stats'), icons.box),
             cmd('app', 'Console', 'Go to console', nav('/console'), icons.terminal),
             cmd('app', 'Edit Config', 'Edit compose-farm.yaml', nav('/console#editor'), icons.file_code),
             cmd('app', 'Docs', 'Open documentation', openExternal('https://compose-farm.nijho.lt/'), icons.book_open),
+            cmd('app', 'GitHub Repo', 'Open GitHub repository', openExternal('https://github.com/basnijholt/compose-farm'), icons.external_link),
         ];
 
         // Add stack-specific actions if on a stack page
@@ -713,11 +744,6 @@ function playFabIntro() {
         input.focus();
     }
 
-    function close() {
-        dialog.close();
-        restoreTheme();
-    }
-
     function exec() {
         const cmd = filtered[selected];
         if (cmd) {
@@ -839,6 +865,119 @@ function initPage() {
     initMonacoEditors();
     initSaveButton();
     updateShortcutKeys();
+    initLiveStats();
+    initSharedActionMenu();
+    maybeRunStackAction();
+}
+
+function navigateToStack(stack, action = null) {
+    const url = action ? `/stack/${stack}?action=${action}` : `/stack/${stack}`;
+    window.location.href = url;
+}
+
+/**
+ * Initialize shared action menu for container rows
+ */
+function initSharedActionMenu() {
+    const menuEl = document.getElementById('shared-action-menu');
+    if (!menuEl) return;
+    if (menuEl.dataset.bound === '1') return;
+    menuEl.dataset.bound = '1';
+
+    let hoverTimeout = null;
+
+    function showMenuForButton(btn, stack) {
+        menuEl.dataset.stack = stack;
+
+        // Position menu relative to button
+        const rect = btn.getBoundingClientRect();
+        menuEl.classList.remove('hidden');
+        menuEl.style.visibility = 'hidden';
+        const menuRect = menuEl.getBoundingClientRect();
+
+        const left = rect.right - menuRect.width + window.scrollX;
+        const top = rect.bottom + window.scrollY;
+
+        menuEl.style.top = `${top}px`;
+        menuEl.style.left = `${left}px`;
+        menuEl.style.visibility = '';
+
+        if (typeof liveStats !== 'undefined') liveStats.dropdownOpen = true;
+    }
+
+    function closeMenu() {
+        menuEl.classList.add('hidden');
+        if (typeof liveStats !== 'undefined') liveStats.dropdownOpen = false;
+        menuEl.dataset.stack = '';
+    }
+
+    function scheduleClose() {
+        if (hoverTimeout) clearTimeout(hoverTimeout);
+        hoverTimeout = setTimeout(closeMenu, 100);
+    }
+
+    function cancelClose() {
+        if (hoverTimeout) {
+            clearTimeout(hoverTimeout);
+            hoverTimeout = null;
+        }
+    }
+
+    // Button hover: show menu (event delegation on tbody)
+    const tbody = document.getElementById('container-rows');
+    if (tbody) {
+        tbody.addEventListener('mouseenter', (e) => {
+            const btn = e.target.closest('button[onclick^="openActionMenu"]');
+            if (!btn) return;
+
+            // Extract stack from onclick attribute
+            const match = btn.getAttribute('onclick')?.match(/openActionMenu\(event,\s*'([^']+)'\)/);
+            if (!match) return;
+
+            cancelClose();
+            showMenuForButton(btn, match[1]);
+        }, true);
+
+        tbody.addEventListener('mouseleave', (e) => {
+            const btn = e.target.closest('button[onclick^="openActionMenu"]');
+            if (btn) scheduleClose();
+        }, true);
+    }
+
+    // Keep menu open while hovering over it
+    menuEl.addEventListener('mouseenter', cancelClose);
+    menuEl.addEventListener('mouseleave', scheduleClose);
+
+    // Click action in menu
+    menuEl.addEventListener('click', (e) => {
+        const link = e.target.closest('a[data-action]');
+        const stack = menuEl.dataset.stack;
+        if (!link || !stack) return;
+
+        e.preventDefault();
+        navigateToStack(stack, link.dataset.action);
+        closeMenu();
+    });
+
+    // Also support click on button (for touch/accessibility)
+    window.openActionMenu = function(event, stack) {
+        event.stopPropagation();
+        showMenuForButton(event.currentTarget, stack);
+    };
+
+    // Close on outside click
+    document.body.addEventListener('click', (e) => {
+        if (!menuEl.classList.contains('hidden') &&
+            !menuEl.contains(e.target) &&
+            !e.target.closest('button[onclick^="openActionMenu"]')) {
+            closeMenu();
+        }
+    });
+
+    // Close on Escape
+    document.body.addEventListener('keydown', (e) => {
+        if (e.key === 'Escape') closeMenu();
+    });
 }
 
 /**
@@ -859,6 +998,30 @@ function tryReconnectToTask(path) {
     });
 }
 
+function maybeRunStackAction() {
+    const params = new URLSearchParams(window.location.search);
+    const stackEl = document.querySelector('[data-stack-name]');
+    const stackName = stackEl?.dataset?.stackName;
+    if (!stackName) return;
+
+    const action = params.get('action');
+    if (!action) return;
+
+    const button = document.querySelector(`button[hx-post="/api/stack/${stackName}/${action}"]`);
+    if (!button) return;
+
+    params.delete('action');
+    const newQuery = params.toString();
+    const newUrl = newQuery ? `${window.location.pathname}?${newQuery}` : window.location.pathname;
+    history.replaceState({}, '', newUrl);
+
+    if (window.htmx) {
+        htmx.trigger(button, 'click');
+    } else {
+        button.click();
+    }
+}
+
 // Initialize on page load
 document.addEventListener('DOMContentLoaded', function() {
     initPage();
@@ -900,3 +1063,443 @@ document.body.addEventListener('htmx:afterRequest', function(evt) {
         // Not valid JSON, ignore
     }
 });
+
+// ============================================================================
+// LIVE STATS PAGE
+// ============================================================================
+
+// State persists across SPA navigation (intervals must be cleared on re-init)
+let liveStats = {
+    sortCol: 9,
+    sortAsc: false,
+    lastUpdate: 0,
+    dropdownOpen: false,
+    scrolling: false,
+    scrollTimer: null,
+    loadingHosts: new Set(),
+    eventsBound: false,
+    intervals: [],
+    updateCheckTimes: new Map(),
+    autoRefresh: true
+};
+
+const REFRESH_INTERVAL = 5000;
+const UPDATE_CHECK_TTL = 120000;
+const NUMERIC_COLS = new Set([8, 9, 10, 11]);  // uptime, cpu, mem, net
+
+function filterTable() {
+    const textFilter = document.getElementById('filter-input')?.value.toLowerCase() || '';
+    const hostFilter = document.getElementById('host-filter')?.value || '';
+    const rows = document.querySelectorAll('#container-rows tr');
+    let visible = 0;
+    let total = 0;
+
+    rows.forEach(row => {
+        // Skip loading/empty/error rows (they have colspan)
+        if (row.cells[0]?.colSpan > 1) return;
+        total++;
+        const matchesText = !textFilter || row.textContent.toLowerCase().includes(textFilter);
+        const matchesHost = !hostFilter || row.dataset.host === hostFilter;
+        const show = matchesText && matchesHost;
+        row.style.display = show ? '' : 'none';
+        if (show) visible++;
+    });
+
+    const countEl = document.getElementById('container-count');
+    if (countEl) {
+        const isFiltering = textFilter || hostFilter;
+        countEl.textContent = total > 0
+            ? (isFiltering ? `${visible} of ${total} containers` : `${total} containers`)
+            : '';
+    }
+}
+window.filterTable = filterTable;
+
+function sortTable(col) {
+    if (liveStats.sortCol === col) {
+        liveStats.sortAsc = !liveStats.sortAsc;
+    } else {
+        liveStats.sortCol = col;
+        liveStats.sortAsc = false;
+    }
+    updateSortIndicators();
+    doSort();
+}
+window.sortTable = sortTable;
+
+function updateSortIndicators() {
+    document.querySelectorAll('thead th').forEach((th, i) => {
+        const span = th.querySelector('.sort-indicator');
+        if (span) {
+            span.textContent = (i === liveStats.sortCol) ? (liveStats.sortAsc ? '↑' : '↓') : '';
+            span.style.opacity = (i === liveStats.sortCol) ? '1' : '0.3';
+        }
+    });
+}
+
+function doSort() {
+    const tbody = document.getElementById('container-rows');
+    if (!tbody) return;
+
+    const rows = Array.from(tbody.querySelectorAll('tr'));
+    if (rows.length === 0) return;
+    if (rows.length === 1 && rows[0].cells[0]?.colSpan > 1) return;  // Empty state row
+
+    const isNumeric = NUMERIC_COLS.has(liveStats.sortCol);
+    rows.sort((a, b) => {
+        // Pin placeholders/empty rows to the bottom
+        const aLoading = a.classList.contains('loading-row') || a.classList.contains('host-empty') || a.cells[0]?.colSpan > 1;
+        const bLoading = b.classList.contains('loading-row') || b.classList.contains('host-empty') || b.cells[0]?.colSpan > 1;
+        if (aLoading && !bLoading) return 1;
+        if (!aLoading && bLoading) return -1;
+        if (aLoading && bLoading) return 0;
+
+        const aVal = a.cells[liveStats.sortCol]?.dataset?.sort ?? '';
+        const bVal = b.cells[liveStats.sortCol]?.dataset?.sort ?? '';
+        const cmp = isNumeric ? aVal - bVal : aVal.localeCompare(bVal);
+        return liveStats.sortAsc ? cmp : -cmp;
+    });
+
+    let index = 1;
+    const fragment = document.createDocumentFragment();
+    rows.forEach((row) => {
+        if (row.cells.length > 1) {
+            row.cells[0].textContent = index++;
+        }
+        fragment.appendChild(row);
+    });
+    tbody.appendChild(fragment);
+}
+
+function isLoading() {
+    return liveStats.loadingHosts.size > 0;
+}
+
+function getLiveStatsHosts() {
+    const tbody = document.getElementById('container-rows');
+    if (!tbody) return [];
+    const dataHosts = tbody.dataset.hosts || '';
+    return dataHosts.split(',').map(h => h.trim()).filter(Boolean);
+}
+
+function buildHostRow(host, message, className) {
+    return (
+        `<tr class="${className}" data-host="${host}">` +
+        `<td colspan="12" class="text-center py-2">` +
+        `<span class="text-sm opacity-60">${message}</span>` +
+        `</td></tr>`
+    );
+}
+
+async function checkUpdatesForHost(host) {
+    // Update checks always run - they only update small cells, not disruptive
+    const last = liveStats.updateCheckTimes.get(host) || 0;
+    if (Date.now() - last < UPDATE_CHECK_TTL) return;
+
+    const cells = Array.from(
+        document.querySelectorAll(`tr[data-host="${host}"] td.update-cell[data-image][data-tag]`)
+    );
+    if (cells.length === 0) return;
+
+    const items = [];
+    const seen = new Set();
+    cells.forEach(cell => {
+        const image = decodeURIComponent(cell.dataset.image || '');
+        const tag = decodeURIComponent(cell.dataset.tag || '');
+        const key = `${image}:${tag}`;
+        if (!image || seen.has(key)) return;
+        seen.add(key);
+        items.push({ image, tag });
+    });
+
+    if (items.length === 0) return;
+
+    try {
+        const response = await fetch('/api/containers/check-updates', {
+            method: 'POST',
+            headers: { 'Content-Type': 'application/json' },
+            body: JSON.stringify({ items })
+        });
+        if (!response.ok) return;
+        const data = await response.json();
+        const results = Array.isArray(data?.results) ? data.results : [];
+        const htmlMap = new Map();
+        results.forEach(result => {
+            const key = `${result.image}:${result.tag}`;
+            htmlMap.set(key, result.html);
+        });
+
+        cells.forEach(cell => {
+            const image = decodeURIComponent(cell.dataset.image || '');
+            const tag = decodeURIComponent(cell.dataset.tag || '');
+            const key = `${image}:${tag}`;
+            const html = htmlMap.get(key);
+            if (html && cell.innerHTML !== html) {
+                cell.innerHTML = html;
+            }
+        });
+
+        liveStats.updateCheckTimes.set(host, Date.now());
+    } catch (e) {
+        console.error('Update check failed:', e);
+    }
+}
+
+function replaceHostRows(host, html) {
+    const tbody = document.getElementById('container-rows');
+    if (!tbody) return;
+
+    // Remove loading indicator for this host if present
+    const loadingRow = tbody.querySelector(`tr.loading-row[data-host="${host}"]`);
+    if (loadingRow) loadingRow.remove();
+
+    const template = document.createElement('template');
+    template.innerHTML = html.trim();
+    let newRows = Array.from(template.content.children).filter(el => el.tagName === 'TR');
+
+    if (newRows.length === 0) {
+        // Only show empty message if we don't have any rows for this host
+        const existing = tbody.querySelector(`tr[data-host="${host}"]:not(.loading-row)`);
+        if (!existing) {
+            template.innerHTML = buildHostRow(host, `No containers on ${host}`, 'host-empty');
+            newRows = Array.from(template.content.children);
+        }
+    }
+
+    // Track which IDs we've seen in this update
+    const newIds = new Set();
+
+    newRows.forEach(newRow => {
+        const id = newRow.id;
+        if (id) newIds.add(id);
+
+        if (id) {
+            const existing = document.getElementById(id);
+            if (existing) {
+                // Morph in place if Idiomorph is available, otherwise replace
+                if (typeof Idiomorph !== 'undefined') {
+                    Idiomorph.morph(existing, newRow);
+                } else {
+                    existing.replaceWith(newRow);
+                }
+
+                // Re-process HTMX if needed (though inner content usually carries attributes)
+                const morphedRow = document.getElementById(id);
+                if (window.htmx) htmx.process(morphedRow);
+
+                // Trigger refresh animation
+                if (morphedRow) {
+                    morphedRow.classList.add('row-updated');
+                    setTimeout(() => morphedRow.classList.remove('row-updated'), 500);
+                }
+            } else {
+                // New row - append (will be sorted later)
+                tbody.appendChild(newRow);
+                if (window.htmx) htmx.process(newRow);
+                // Animate new rows too
+                newRow.classList.add('row-updated');
+                setTimeout(() => newRow.classList.remove('row-updated'), 500);
+            }
+        } else {
+            // Fallback for rows without ID (like error/empty messages)
+            // Just append them, cleaning up previous generic rows handled below
+            tbody.appendChild(newRow);
+        }
+    });
+
+    // Remove orphaned rows for this host (rows that exist in DOM but not in new response)
+    // Be careful not to remove rows that were just added (if they lack IDs)
+    const currentHostRows = Array.from(tbody.querySelectorAll(`tr[data-host="${host}"]`));
+    currentHostRows.forEach(row => {
+        // Skip if it's one of the new rows we just appended (check presence in newRows?)
+        // Actually, if we just appended it, it is in DOM.
+        // We rely on ID matching.
+        // Error/Empty rows usually don't have ID, but we handle them by clearing old ones?
+        // Let's assume data rows have IDs.
+        if (row.id && !newIds.has(row.id)) {
+            row.remove();
+        }
+        // Also remove old empty/error messages if we now have data
+        if (!row.id && newRows.length > 0 && newRows[0].id) {
+             row.remove();
+        }
+    });
+
+    liveStats.loadingHosts.delete(host);
+    checkUpdatesForHost(host);
+    scheduleRowUpdate();
+}
+
+async function loadHostRows(host) {
+    liveStats.loadingHosts.add(host);
+    try {
+        const response = await fetch(`/api/containers/rows/${encodeURIComponent(host)}`);
+        const html = response.ok ? await response.text() : '';
+        replaceHostRows(host, html);
+    } catch (e) {
+        console.error(`Failed to load ${host}:`, e);
+        const msg = e.message || String(e);
+        // Fallback to simpler error display if replaceHostRows fails (e.g. Idiomorph missing)
+        try {
+            replaceHostRows(host, buildHostRow(host, `Error: ${msg}`, 'text-error'));
+        } catch (err2) {
+            // Last resort: find row and force innerHTML
+            const tbody = document.getElementById('container-rows');
+            const row = tbody?.querySelector(`tr[data-host="${host}"]`);
+            if (row) row.innerHTML = `<td colspan="12" class="text-center text-error">Error: ${msg}</td>`;
+        }
+    } finally {
+        liveStats.loadingHosts.delete(host);
+    }
+}
+
+function refreshLiveStats() {
+    if (liveStats.dropdownOpen || liveStats.scrolling) return;
+    const hosts = getLiveStatsHosts();
+    if (hosts.length === 0) return;
+    liveStats.lastUpdate = Date.now();
+    hosts.forEach(loadHostRows);
+}
+window.refreshLiveStats = refreshLiveStats;
+
+function toggleAutoRefresh() {
+    liveStats.autoRefresh = !liveStats.autoRefresh;
+    const timer = document.getElementById('refresh-timer');
+    if (timer) {
+        timer.classList.toggle('btn-error', !liveStats.autoRefresh);
+        timer.classList.toggle('btn-outline', liveStats.autoRefresh);
+    }
+    if (liveStats.autoRefresh) {
+        // Re-enabling: trigger immediate refresh
+        refreshLiveStats();
+    } else {
+        // Disabling: ensure update checks run for current data
+        const hosts = getLiveStatsHosts();
+        hosts.forEach(host => checkUpdatesForHost(host));
+    }
+}
+window.toggleAutoRefresh = toggleAutoRefresh;
+
+function initLiveStats() {
+    if (!document.getElementById('refresh-timer')) return;
+
+    // Clear previous intervals (important for SPA navigation)
+    liveStats.intervals.forEach(clearInterval);
+    liveStats.intervals = [];
+    liveStats.lastUpdate = Date.now();
+    liveStats.dropdownOpen = false;
+    liveStats.scrolling = false;
+    if (liveStats.scrollTimer) clearTimeout(liveStats.scrollTimer);
+    liveStats.scrollTimer = null;
+    liveStats.loadingHosts.clear();
+    liveStats.updateCheckTimes = new Map();
+    liveStats.autoRefresh = true;
+
+    if (!liveStats.eventsBound) {
+        liveStats.eventsBound = true;
+
+        // Dropdown pauses refresh
+        document.body.addEventListener('click', e => {
+            liveStats.dropdownOpen = !!e.target.closest('.dropdown');
+        });
+        document.body.addEventListener('focusin', e => {
+            if (e.target.closest('.dropdown')) liveStats.dropdownOpen = true;
+        });
+        document.body.addEventListener('focusout', () => {
+            setTimeout(() => {
+                liveStats.dropdownOpen = !!document.activeElement?.closest('.dropdown');
+            }, 150);
+        });
+        document.body.addEventListener('keydown', e => {
+            if (e.key === 'Escape') liveStats.dropdownOpen = false;
+        });
+
+        // Pause refresh while scrolling (helps on slow mobile browsers)
+        window.addEventListener('scroll', () => {
+            liveStats.scrolling = true;
+            if (liveStats.scrollTimer) clearTimeout(liveStats.scrollTimer);
+            liveStats.scrollTimer = setTimeout(() => {
+                liveStats.scrolling = false;
+            }, 200);
+        }, { passive: true });
+    }
+
+    // Auto-refresh every 5 seconds (skip if disabled, loading, or dropdown open)
+    liveStats.intervals.push(setInterval(() => {
+        if (!liveStats.autoRefresh) return;
+        if (liveStats.dropdownOpen || liveStats.scrolling || isLoading()) return;
+        refreshLiveStats();
+    }, REFRESH_INTERVAL));
+
+    // Timer display (updates every 100ms)
+    liveStats.intervals.push(setInterval(() => {
+        const timer = document.getElementById('refresh-timer');
+        if (!timer) {
+            liveStats.intervals.forEach(clearInterval);
+            return;
+        }
+
+        const loading = isLoading();
+        const paused = liveStats.dropdownOpen || liveStats.scrolling;
+        const elapsed = Date.now() - liveStats.lastUpdate;
+        window.refreshPaused = paused || loading || !liveStats.autoRefresh;
+
+        // Update refresh timer button
+        let text;
+        if (!liveStats.autoRefresh) {
+            text = 'OFF';
+        } else if (paused) {
+            text = '❚❚';
+        } else {
+            const remaining = Math.max(0, REFRESH_INTERVAL - elapsed);
+            text = loading ? '↻ …' : `↻ ${Math.ceil(remaining / 1000)}s`;
+        }
+        if (timer.textContent !== text) {
+            timer.textContent = text;
+        }
+
+        // Update "last updated" display
+        const lastUpdatedEl = document.getElementById('last-updated');
+        if (lastUpdatedEl) {
+            const secs = Math.floor(elapsed / 1000);
+            const updatedText = secs < 5 ? 'Updated just now' : `Updated ${secs}s ago`;
+            if (lastUpdatedEl.textContent !== updatedText) {
+                lastUpdatedEl.textContent = updatedText;
+            }
+        }
+    }, 100));
+
+    updateSortIndicators();
+    refreshLiveStats();
+}
+
+function scheduleRowUpdate() {
+    // Sort and filter immediately to prevent flicker
+    doSort();
+    filterTable();
+}
+
+// ============================================================================
+// STACKS BY HOST FILTER
+// ============================================================================
+
+function sbhFilter() {
+    const query = (document.getElementById('sbh-filter')?.value || '').toLowerCase();
+    const hostFilter = document.getElementById('sbh-host-select')?.value || '';
+
+    document.querySelectorAll('.sbh-group').forEach(group => {
+        if (hostFilter && group.dataset.h !== hostFilter) {
+            group.hidden = true;
+            return;
+        }
+
+        let visibleCount = 0;
+        group.querySelectorAll('li[data-s]').forEach(li => {
+            const show = !query || li.dataset.s.includes(query);
+            li.hidden = !show;
+            if (show) visibleCount++;
+        });
+        group.hidden = visibleCount === 0;
+    });
+}
+window.sbhFilter = sbhFilter;

src/compose_farm/web/templates/base.html

@@ -26,6 +26,23 @@
     </script>
 </head>
 <body class="min-h-screen bg-base-200">
+    <svg style="display: none">
+        <symbol id="icon-menu" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+            <circle cx="12" cy="5" r="1" /><circle cx="12" cy="12" r="1" /><circle cx="12" cy="19" r="1" />
+        </symbol>
+        <symbol id="icon-restart" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+            <path d="M4 4v5h.582m15.356 2A8.001 8.001 0 004.582 9m0 0H9m11 11v-5h-.581m0 0a8.003 8.003 0 01-15.357-2m15.357 2H15" />
+        </symbol>
+        <symbol id="icon-pull" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+            <path d="M4 16v1a3 3 0 003 3h10a3 3 0 003-3v-1m-4-4l-4 4m0 0l-4-4m4 4V4" />
+        </symbol>
+        <symbol id="icon-update" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+            <path d="M4 16v1a3 3 0 003 3h10a3 3 0 003-3v-1m-4-8l-4-4m0 0L8 8m4-4v12" />
+        </symbol>
+        <symbol id="icon-logs" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+            <path d="M9 12h6m-6 4h6m2 5H7a2 2 0 01-2-2V5a2 2 0 012-2h5.586a1 1 0 01.707.293l5.414 5.414a1 1 0 01.293.707V19a2 2 0 01-2 2z" />
+        </symbol>
+    </svg>
     <div class="drawer lg:drawer-open">
         <input id="drawer-toggle" type="checkbox" class="drawer-toggle" />
 
@@ -80,6 +97,8 @@
 
     <!-- Scripts - HTMX first -->
     <script src="https://unpkg.com/htmx.org@2.0.4" data-vendor="htmx.js"></script>
+    <script src="https://unpkg.com/idiomorph/dist/idiomorph.min.js"></script>
+    <script src="https://unpkg.com/idiomorph/dist/idiomorph-ext.min.js"></script>
     <script src="https://cdn.jsdelivr.net/npm/@xterm/xterm@5.5.0/lib/xterm.js" data-vendor="xterm.js"></script>
     <script src="https://cdn.jsdelivr.net/npm/@xterm/addon-fit@0.10.0/lib/addon-fit.js" data-vendor="xterm-fit.js"></script>
     <script src="/static/app.js"></script>

src/compose_farm/web/templates/containers.html

@@ -0,0 +1,97 @@
+{% extends "base.html" %}
+{% from "partials/components.html" import page_header %}
+{% from "partials/icons.html" import refresh_cw %}
+{% block title %}Live Stats - Compose Farm{% endblock %}
+
+{% block content %}
+<div class="max-w-7xl">
+    {{ page_header("Live Stats", "All running containers across hosts") }}
+
+    {% if not glances_enabled %}
+    <div class="alert alert-warning mb-6">
+        <svg xmlns="http://www.w3.org/2000/svg" class="stroke-current shrink-0 h-6 w-6" fill="none" viewBox="0 0 24 24"><path stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M12 9v2m0 4h.01m-6.938 4h13.856c1.54 0 2.502-1.667 1.732-3L13.732 4c-.77-1.333-2.694-1.333-3.464 0L3.34 16c-.77 1.333.192 3 1.732 3z" /></svg>
+        <div>
+            <h3 class="font-bold">Glances not configured</h3>
+            <div class="text-xs">Add <code class="bg-base-300 px-1 rounded">glances_stack: glances</code> to your config and deploy Glances on all hosts.</div>
+        </div>
+    </div>
+    {% else %}
+
+    <!-- Action Bar -->
+    <div class="flex flex-wrap items-center gap-4 mb-6">
+        <div class="tooltip" data-tip="Refresh now">
+            <button class="btn btn-outline btn-sm" type="button" onclick="refreshLiveStats()">
+                {{ refresh_cw() }} Refresh
+            </button>
+        </div>
+        <div class="tooltip" data-tip="Click to toggle auto-refresh">
+            <button class="btn btn-outline btn-sm font-mono w-20 justify-center"
+                    id="refresh-timer" onclick="toggleAutoRefresh()">↻</button>
+        </div>
+        <input type="text" id="filter-input" placeholder="Filter containers..."
+               class="input input-bordered input-sm w-64" onkeyup="filterTable()">
+        <select id="host-filter" class="select select-bordered select-sm" onchange="filterTable()">
+            <option value="">All hosts</option>
+            {% for host in hosts %}<option value="{{ host }}">{{ host }}</option>{% endfor %}
+        </select>
+        <span id="container-count" class="text-sm text-base-content/60"></span>
+        <span id="last-updated" class="text-sm text-base-content/40 ml-auto"></span>
+    </div>
+
+    <!-- Container Table -->
+    <div class="card bg-base-100 shadow overflow-x-auto">
+        <table class="table table-zebra table-sm w-full">
+            <thead class="sticky top-0 bg-base-200">
+                <tr>
+                    <th class="w-8">#</th>
+                    <th class="cursor-pointer" onclick="sortTable(1)">Stack<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer" onclick="sortTable(2)">Service<span class="sort-indicator"></span></th>
+                    <th></th>
+                    <th class="cursor-pointer" onclick="sortTable(4)">Host<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer" onclick="sortTable(5)">Image<span class="sort-indicator"></span></th>
+                    <th class="w-16">Update</th>
+                    <th class="cursor-pointer" onclick="sortTable(7)">Status<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer text-right" onclick="sortTable(8)">Uptime<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer text-right" onclick="sortTable(9)">CPU<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer text-right" onclick="sortTable(10)">Mem<span class="sort-indicator"></span></th>
+                    <th class="cursor-pointer text-right" onclick="sortTable(11)">Net I/O<span class="sort-indicator"></span></th>
+                </tr>
+            </thead>
+            <tbody id="container-rows" data-hosts="{{ hosts | join(',') }}">
+                {% for host in hosts %}
+                <tr class="loading-row" data-host="{{ host }}">
+                    <td colspan="12" class="text-center py-2">
+                        <span class="loading loading-spinner loading-xs"></span>
+                        <span class="text-sm opacity-60">Loading {{ host }}...</span>
+                    </td>
+                </tr>
+                {% endfor %}
+            </tbody>
+        </table>
+    </div>
+    {% endif %}
+
+    <!-- Shared Action Menu -->
+    <ul id="shared-action-menu" class="menu menu-sm bg-base-200 rounded-box shadow-lg w-36 absolute z-50 p-2 hidden">
+        <li><a data-action="restart"><svg class="h-4 w-4"><use href="#icon-restart" /></svg>Restart</a></li>
+        <li><a data-action="pull"><svg class="h-4 w-4"><use href="#icon-pull" /></svg>Pull</a></li>
+        <li><a data-action="update"><svg class="h-4 w-4"><use href="#icon-update" /></svg>Update</a></li>
+        <li><a data-action="logs"><svg class="h-4 w-4"><use href="#icon-logs" /></svg>Logs</a></li>
+    </ul>
+</div>
+{% endblock %}
+
+{% block scripts %}
+{% if glances_enabled %}
+<style>
+.sort-indicator { display: inline-block; width: 1em; text-align: center; opacity: 0.5; }
+.high-usage { background-color: oklch(var(--er) / 0.15) !important; }
+/* Refresh animation */
+@keyframes row-pulse {
+    0% { background-color: oklch(var(--p) / 0.2); }
+    100% { background-color: transparent; }
+}
+.row-updated { animation: row-pulse 0.5s ease-out; }
+</style>
+{% endif %}
+{% endblock %}

src/compose_farm/web/templates/index.html

@@ -53,6 +53,13 @@
         {% include "partials/stacks_by_host.html" %}
     </div>
 
+    <!-- Host Resources (Glances) -->
+    <div id="glances-stats"
+         hx-get="/api/glances"
+         hx-trigger="load, cf:refresh from:body, every 30s"
+         hx-swap="innerHTML">
+    </div>
+
     <!-- Hosts Configuration -->
     {% call collapse("Hosts (" ~ (hosts | length) ~ ")", icon=server()) %}
         {% call table() %}

src/compose_farm/web/templates/partials/glances.html

@@ -0,0 +1,66 @@
+{# Glances resource stats display #}
+{% from "partials/icons.html" import cpu, memory_stick, gauge, server, activity, hard_drive, arrow_down_up, refresh_cw %}
+
+{% macro progress_bar(percent, color="primary") %}
+<div class="flex items-center gap-2 min-w-32">
+    <progress class="progress progress-{{ color }} flex-1" value="{{ percent }}" max="100"></progress>
+    <span class="text-xs w-10 text-right">{{ "%.1f"|format(percent) }}%</span>
+</div>
+{% endmacro %}
+
+{% macro format_rate(bytes_per_sec) %}
+{%- if bytes_per_sec >= 1048576 -%}
+{{ "%.1f"|format(bytes_per_sec / 1048576) }} MB/s
+{%- elif bytes_per_sec >= 1024 -%}
+{{ "%.1f"|format(bytes_per_sec / 1024) }} KB/s
+{%- else -%}
+{{ "%.0f"|format(bytes_per_sec) }} B/s
+{%- endif -%}
+{% endmacro %}
+
+{% macro host_row(host_stats) %}
+<tr>
+    <td class="font-medium">{{ server(14) }} {{ host_stats.host }}</td>
+    {% if host_stats.error %}
+    <td colspan="5" class="text-error text-xs">{{ host_stats.error }}</td>
+    {% else %}
+    <td>{{ progress_bar(host_stats.cpu_percent, "info") }}</td>
+    <td>{{ progress_bar(host_stats.mem_percent, "success") }}</td>
+    <td>{{ progress_bar(host_stats.disk_percent, "warning") }}</td>
+    <td class="text-xs font-mono">↓{{ format_rate(host_stats.net_rx_rate) }} ↑{{ format_rate(host_stats.net_tx_rate) }}</td>
+    <td class="text-sm">{{ "%.1f"|format(host_stats.load) }}</td>
+    {% endif %}
+</tr>
+{% endmacro %}
+
+<div class="card bg-base-100 shadow mt-4 mb-4">
+    <div class="card-body p-4">
+        <div class="flex items-center justify-between">
+            <h2 class="card-title text-base gap-2">{{ activity(18) }} Host Resources</h2>
+            <button class="btn btn-ghost btn-xs opacity-50 hover:opacity-100"
+                    hx-get="/api/glances" hx-target="#glances-stats" hx-swap="innerHTML"
+                    title="Refresh">
+                {{ refresh_cw(14) }}
+            </button>
+        </div>
+        <div class="overflow-x-auto">
+            <table class="table table-sm">
+                <thead>
+                    <tr>
+                        <th>Host</th>
+                        <th>{{ cpu(14) }} CPU</th>
+                        <th>{{ memory_stick(14) }} Memory</th>
+                        <th>{{ hard_drive(14) }} Disk</th>
+                        <th>{{ arrow_down_up(14) }} Net</th>
+                        <th>{{ gauge(14) }} Load</th>
+                    </tr>
+                </thead>
+                <tbody>
+                    {% for host_name, host_stats in stats.items() %}
+                    {{ host_row(host_stats) }}
+                    {% endfor %}
+                </tbody>
+            </table>
+        </div>
+    </div>
+</div>

src/compose_farm/web/templates/partials/icons.html

@@ -176,3 +176,46 @@
     <path d="M15 3h6v6"/><path d="M10 14 21 3"/><path d="M18 13v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/>
 </svg>
 {% endmacro %}
+
+{# Resource monitoring icons #}
+{% macro cpu(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <rect width="16" height="16" x="4" y="4" rx="2"/><rect width="6" height="6" x="9" y="9" rx="1"/><path d="M15 2v2"/><path d="M15 20v2"/><path d="M2 15h2"/><path d="M2 9h2"/><path d="M20 15h2"/><path d="M20 9h2"/><path d="M9 2v2"/><path d="M9 20v2"/>
+</svg>
+{% endmacro %}
+
+{% macro memory_stick(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M6 19v-3"/><path d="M10 19v-3"/><path d="M14 19v-3"/><path d="M18 19v-3"/><path d="M8 11V9"/><path d="M16 11V9"/><path d="M12 11V9"/><path d="M2 15h20"/><path d="M2 7a2 2 0 0 1 2-2h16a2 2 0 0 1 2 2v1.1a2 2 0 0 0 0 3.837V17a2 2 0 0 1-2 2H4a2 2 0 0 1-2-2v-5.1a2 2 0 0 0 0-3.837z"/>
+</svg>
+{% endmacro %}
+
+{% macro gauge(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="m12 14 4-4"/><path d="M3.34 19a10 10 0 1 1 17.32 0"/>
+</svg>
+{% endmacro %}
+
+{% macro activity(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M22 12h-2.48a2 2 0 0 0-1.93 1.46l-2.35 8.36a.25.25 0 0 1-.48 0L9.24 2.18a.25.25 0 0 0-.48 0l-2.35 8.36A2 2 0 0 1 4.49 12H2"/>
+</svg>
+{% endmacro %}
+
+{% macro arrow_down_up(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="m3 16 4 4 4-4"/><path d="M7 20V4"/><path d="m21 8-4-4-4 4"/><path d="M17 4v16"/>
+</svg>
+{% endmacro %}
+
+{% macro hard_drive(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <line x1="22" x2="2" y1="12" y2="12"/><path d="M5.45 5.11 2 12v6a2 2 0 0 0 2 2h16a2 2 0 0 0 2-2v-6l-3.45-6.89A2 2 0 0 0 16.76 4H7.24a2 2 0 0 0-1.79 1.11z"/><line x1="6" x2="6.01" y1="16" y2="16"/><line x1="10" x2="10.01" y1="16" y2="16"/>
+</svg>
+{% endmacro %}
+
+{% macro box(size=16) %}
+<svg xmlns="http://www.w3.org/2000/svg" width="{{ size }}" height="{{ size }}" viewBox="0 0 24 24" fill="none" stroke="currentColor" stroke-width="2" stroke-linecap="round" stroke-linejoin="round">
+    <path d="M21 8a2 2 0 0 0-1-1.73l-7-4a2 2 0 0 0-2 0l-7 4A2 2 0 0 0 3 8v8a2 2 0 0 0 1 1.73l7 4a2 2 0 0 0 2 0l7-4A2 2 0 0 0 21 16Z"/><path d="m3.3 7 8.7 5 8.7-5"/><path d="M12 22V12"/>
+</svg>
+{% endmacro %}

src/compose_farm/web/templates/partials/sidebar.html

@@ -1,8 +1,9 @@
-{% from "partials/icons.html" import home, search, terminal %}
+{% from "partials/icons.html" import home, search, terminal, box %}
 <!-- Navigation Links -->
 <div class="mb-4">
     <ul class="menu" hx-boost="true" hx-target="#main-content" hx-select="#main-content" hx-swap="outerHTML">
         <li><a href="/" class="font-semibold">{{ home() }} Dashboard</a></li>
+        <li><a href="/live-stats" class="font-semibold">{{ box() }} Live Stats</a></li>
         <li><a href="/console" class="font-semibold">{{ terminal() }} Console</a></li>
     </ul>
 </div>

src/compose_farm/web/templates/partials/stacks_by_host.html

@@ -20,20 +20,4 @@
     {% else %}
     <p class="text-base-content/60 italic">No stacks currently running.</p>
     {% endfor %}
-    <script>
-    function sbhFilter() {
-        const q = (document.getElementById('sbh-filter')?.value || '').toLowerCase();
-        const h = document.getElementById('sbh-host-select')?.value || '';
-        document.querySelectorAll('.sbh-group').forEach(g => {
-            if (h && g.dataset.h !== h) { g.hidden = true; return; }
-            let n = 0;
-            g.querySelectorAll('li[data-s]').forEach(li => {
-                const show = !q || li.dataset.s.includes(q);
-                li.hidden = !show;
-                if (show) n++;
-            });
-            g.hidden = !n;
-        });
-    }
-    </script>
 {% endcall %}

src/compose_farm/web/templates/stack.html

@@ -4,7 +4,7 @@
 {% block title %}{{ name }} - Compose Farm{% endblock %}
 
 {% block content %}
-<div class="max-w-5xl" data-services="{{ services | join(',') }}" data-containers='{{ containers | tojson }}' data-website-urls='{{ website_urls | tojson }}'>
+<div class="max-w-5xl" data-stack-name="{{ name }}" data-services="{{ services | join(',') }}" data-containers='{{ containers | tojson }}' data-website-urls='{{ website_urls | tojson }}'>
     <div class="mb-6">
         <h1 class="text-3xl font-bold rainbow-hover">{{ name }}</h1>
         <div class="flex flex-wrap items-center gap-2 mt-2">

src/compose_farm/web/ws.py

@@ -37,7 +37,7 @@ def _parse_resize(msg: str) -> tuple[int, int] | None:
     """Parse a resize message, return (cols, rows) or None if not a resize."""
     try:
         data = json.loads(msg)
-        if data.get("type") == "resize":
+        if isinstance(data, dict) and data.get("type") == "resize":
             return int(data["cols"]), int(data["rows"])
     except (json.JSONDecodeError, KeyError, TypeError, ValueError):
         pass

tests/test_cli_lifecycle.py

@@ -58,8 +58,9 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.get_orphaned_stacks", return_value={}),
             patch("compose_farm.cli.lifecycle.get_stacks_needing_migration", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
         ):
-            apply(dry_run=False, no_orphans=False, full=False, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=False, config=None)
 
         captured = capsys.readouterr()
         assert "Nothing to apply" in captured.out
@@ -82,10 +83,11 @@ class TestApplyCommand:
             ),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stack_host", return_value="host1"),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch("compose_farm.cli.lifecycle.stop_orphaned_stacks") as mock_stop,
             patch("compose_farm.cli.lifecycle.up_stacks") as mock_up,
         ):
-            apply(dry_run=True, no_orphans=False, full=False, config=None)
+            apply(dry_run=True, no_orphans=False, no_strays=False, full=False, config=None)
 
         captured = capsys.readouterr()
         assert "Stacks to migrate" in captured.out
@@ -112,6 +114,7 @@ class TestApplyCommand:
             ),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stack_host", return_value="host1"),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -120,7 +123,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.maybe_regenerate_traefik"),
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=False, full=False, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=False, config=None)
 
             mock_up.assert_called_once()
             call_args = mock_up.call_args
@@ -139,6 +142,7 @@ class TestApplyCommand:
             ),
             patch("compose_farm.cli.lifecycle.get_stacks_needing_migration", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -146,7 +150,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.stop_orphaned_stacks") as mock_stop,
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=False, full=False, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=False, config=None)
 
             mock_stop.assert_called_once_with(cfg)
 
@@ -169,6 +173,7 @@ class TestApplyCommand:
             ),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stack_host", return_value="host1"),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -178,7 +183,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.maybe_regenerate_traefik"),
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=True, full=False, config=None)
+            apply(dry_run=False, no_orphans=True, no_strays=False, full=False, config=None)
 
             # Should run migrations but not orphan cleanup
             mock_up.assert_called_once()
@@ -202,8 +207,9 @@ class TestApplyCommand:
             ),
             patch("compose_farm.cli.lifecycle.get_stacks_needing_migration", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
         ):
-            apply(dry_run=False, no_orphans=True, full=False, config=None)
+            apply(dry_run=False, no_orphans=True, no_strays=False, full=False, config=None)
 
         captured = capsys.readouterr()
         assert "Nothing to apply" in captured.out
@@ -221,6 +227,7 @@ class TestApplyCommand:
                 "compose_farm.cli.lifecycle.get_stacks_not_in_state",
                 return_value=["svc1"],
             ),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -229,7 +236,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.maybe_regenerate_traefik"),
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=False, full=False, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=False, config=None)
 
             mock_up.assert_called_once()
             call_args = mock_up.call_args
@@ -249,8 +256,9 @@ class TestApplyCommand:
                 "compose_farm.cli.lifecycle.get_stacks_not_in_state",
                 return_value=["svc1"],
             ),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
         ):
-            apply(dry_run=True, no_orphans=False, full=False, config=None)
+            apply(dry_run=True, no_orphans=False, no_strays=False, full=False, config=None)
 
         captured = capsys.readouterr()
         assert "Stacks to start" in captured.out
@@ -267,6 +275,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.get_orphaned_stacks", return_value={}),
             patch("compose_farm.cli.lifecycle.get_stacks_needing_migration", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -275,7 +284,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.maybe_regenerate_traefik"),
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=False, full=True, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=True, config=None)
 
             mock_up.assert_called_once()
             call_args = mock_up.call_args
@@ -293,8 +302,9 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.get_orphaned_stacks", return_value={}),
             patch("compose_farm.cli.lifecycle.get_stacks_needing_migration", return_value=[]),
             patch("compose_farm.cli.lifecycle.get_stacks_not_in_state", return_value=[]),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
         ):
-            apply(dry_run=True, no_orphans=False, full=True, config=None)
+            apply(dry_run=True, no_orphans=False, no_strays=False, full=True, config=None)
 
         captured = capsys.readouterr()
         assert "Stacks to refresh" in captured.out
@@ -319,6 +329,7 @@ class TestApplyCommand:
                 return_value=["svc2"],
             ),
             patch("compose_farm.cli.lifecycle.get_stack_host", return_value="host2"),
+            patch("compose_farm.cli.lifecycle._discover_strays", return_value={}),
             patch(
                 "compose_farm.cli.lifecycle.run_async",
                 return_value=mock_results,
@@ -327,7 +338,7 @@ class TestApplyCommand:
             patch("compose_farm.cli.lifecycle.maybe_regenerate_traefik"),
             patch("compose_farm.cli.lifecycle.report_results"),
         ):
-            apply(dry_run=False, no_orphans=False, full=True, config=None)
+            apply(dry_run=False, no_orphans=False, no_strays=False, full=True, config=None)
 
             # up_stacks should be called 3 times: migrate, start, refresh
             assert mock_up.call_count == 3

tests/test_containers.py

@@ -0,0 +1,269 @@
+"""Tests for Containers page routes."""
+
+from pathlib import Path
+from unittest.mock import AsyncMock, patch
+
+import pytest
+from fastapi.testclient import TestClient
+
+from compose_farm.config import Config, Host
+from compose_farm.glances import ContainerStats
+from compose_farm.web.app import create_app
+from compose_farm.web.routes.containers import (
+    _format_bytes,
+    _infer_stack_service,
+    _parse_image,
+    _parse_uptime_seconds,
+)
+
+# Byte size constants for tests
+KB = 1024
+MB = KB * 1024
+GB = MB * 1024
+
+
+class TestFormatBytes:
+    """Tests for _format_bytes function (uses humanize library)."""
+
+    def test_bytes(self) -> None:
+        assert _format_bytes(500) == "500 Bytes"
+        assert _format_bytes(0) == "0 Bytes"
+
+    def test_kilobytes(self) -> None:
+        assert _format_bytes(KB) == "1.0 KiB"
+        assert _format_bytes(KB * 5) == "5.0 KiB"
+        assert _format_bytes(KB + 512) == "1.5 KiB"
+
+    def test_megabytes(self) -> None:
+        assert _format_bytes(MB) == "1.0 MiB"
+        assert _format_bytes(MB * 100) == "100.0 MiB"
+        assert _format_bytes(MB * 512) == "512.0 MiB"
+
+    def test_gigabytes(self) -> None:
+        assert _format_bytes(GB) == "1.0 GiB"
+        assert _format_bytes(GB * 2) == "2.0 GiB"
+
+
+class TestParseImage:
+    """Tests for _parse_image function."""
+
+    def test_simple_image_with_tag(self) -> None:
+        assert _parse_image("nginx:latest") == ("nginx", "latest")
+        assert _parse_image("redis:7") == ("redis", "7")
+
+    def test_image_without_tag(self) -> None:
+        assert _parse_image("nginx") == ("nginx", "latest")
+
+    def test_registry_image(self) -> None:
+        assert _parse_image("ghcr.io/user/repo:v1.0") == ("ghcr.io/user/repo", "v1.0")
+        assert _parse_image("docker.io/library/nginx:alpine") == (
+            "docker.io/library/nginx",
+            "alpine",
+        )
+
+    def test_image_with_port_in_registry(self) -> None:
+        # Registry with port should not be confused with tag
+        assert _parse_image("localhost:5000/myimage") == ("localhost:5000/myimage", "latest")
+
+
+class TestParseUptimeSeconds:
+    """Tests for _parse_uptime_seconds function."""
+
+    def test_seconds(self) -> None:
+        assert _parse_uptime_seconds("17 seconds") == 17
+        assert _parse_uptime_seconds("1 second") == 1
+
+    def test_minutes(self) -> None:
+        assert _parse_uptime_seconds("5 minutes") == 300
+        assert _parse_uptime_seconds("1 minute") == 60
+
+    def test_hours(self) -> None:
+        assert _parse_uptime_seconds("2 hours") == 7200
+        assert _parse_uptime_seconds("an hour") == 3600
+        assert _parse_uptime_seconds("1 hour") == 3600
+
+    def test_days(self) -> None:
+        assert _parse_uptime_seconds("3 days") == 259200
+        assert _parse_uptime_seconds("a day") == 86400
+
+    def test_empty(self) -> None:
+        assert _parse_uptime_seconds("") == 0
+        assert _parse_uptime_seconds("-") == 0
+
+
+class TestInferStackService:
+    """Tests for _infer_stack_service function."""
+
+    def test_underscore_separator(self) -> None:
+        assert _infer_stack_service("mystack_web_1") == ("mystack", "web")
+        assert _infer_stack_service("app_db_1") == ("app", "db")
+
+    def test_hyphen_separator(self) -> None:
+        assert _infer_stack_service("mystack-web-1") == ("mystack", "web")
+        assert _infer_stack_service("compose-farm-api-1") == ("compose", "farm-api")
+
+    def test_simple_name(self) -> None:
+        # No separator - use name for both
+        assert _infer_stack_service("nginx") == ("nginx", "nginx")
+        assert _infer_stack_service("traefik") == ("traefik", "traefik")
+
+    def test_single_part_with_separator(self) -> None:
+        # Edge case: separator with empty second part
+        assert _infer_stack_service("single_") == ("single", "")
+
+
+class TestContainersPage:
+    """Tests for containers page endpoint."""
+
+    @pytest.fixture
+    def client(self) -> TestClient:
+        app = create_app()
+        return TestClient(app)
+
+    @pytest.fixture
+    def mock_config(self) -> Config:
+        return Config(
+            compose_dir=Path("/opt/compose"),
+            hosts={
+                "nas": Host(address="192.168.1.6"),
+                "nuc": Host(address="192.168.1.2"),
+            },
+            stacks={"test": "nas"},
+            glances_stack="glances",
+        )
+
+    def test_containers_page_without_glances(self, client: TestClient) -> None:
+        """Test containers page shows warning when Glances not configured."""
+        with patch("compose_farm.web.routes.containers.get_config") as mock:
+            mock.return_value = Config(
+                compose_dir=Path("/opt/compose"),
+                hosts={"nas": Host(address="192.168.1.6")},
+                stacks={"test": "nas"},
+                glances_stack=None,
+            )
+            response = client.get("/live-stats")
+
+        assert response.status_code == 200
+        assert "Glances not configured" in response.text
+
+    def test_containers_page_with_glances(self, client: TestClient, mock_config: Config) -> None:
+        """Test containers page loads when Glances is configured."""
+        with patch("compose_farm.web.routes.containers.get_config") as mock:
+            mock.return_value = mock_config
+            response = client.get("/live-stats")
+
+        assert response.status_code == 200
+        assert "Live Stats" in response.text
+        assert "container-rows" in response.text
+
+
+class TestContainersRowsAPI:
+    """Tests for containers rows HTML endpoint."""
+
+    @pytest.fixture
+    def client(self) -> TestClient:
+        app = create_app()
+        return TestClient(app)
+
+    def test_rows_without_glances(self, client: TestClient) -> None:
+        """Test rows endpoint returns error when Glances not configured."""
+        with patch("compose_farm.web.routes.containers.get_config") as mock:
+            mock.return_value = Config(
+                compose_dir=Path("/opt/compose"),
+                hosts={"nas": Host(address="192.168.1.6")},
+                stacks={"test": "nas"},
+                glances_stack=None,
+            )
+            response = client.get("/api/containers/rows")
+
+        assert response.status_code == 200
+        assert "Glances not configured" in response.text
+
+    def test_rows_returns_html(self, client: TestClient) -> None:
+        """Test rows endpoint returns HTML table rows."""
+        mock_containers = [
+            ContainerStats(
+                name="nginx",
+                host="nas",
+                status="running",
+                image="nginx:latest",
+                cpu_percent=5.5,
+                memory_usage=104857600,
+                memory_limit=1073741824,
+                memory_percent=9.77,
+                network_rx=1000,
+                network_tx=500,
+                uptime="2 hours",
+                ports="80->80/tcp",
+                engine="docker",
+                stack="web",
+                service="nginx",
+            ),
+        ]
+
+        with (
+            patch("compose_farm.web.routes.containers.get_config") as mock_config,
+            patch(
+                "compose_farm.web.routes.containers.fetch_all_container_stats",
+                new_callable=AsyncMock,
+            ) as mock_fetch,
+        ):
+            mock_config.return_value = Config(
+                compose_dir=Path("/opt/compose"),
+                hosts={"nas": Host(address="192.168.1.6")},
+                stacks={"test": "nas"},
+                glances_stack="glances",
+            )
+            mock_fetch.return_value = mock_containers
+
+            response = client.get("/api/containers/rows")
+
+        assert response.status_code == 200
+        assert "<tr " in response.text  # <tr id="..."> has attributes
+        assert "nginx" in response.text
+        assert "running" in response.text
+
+    def test_rows_have_data_sort_attributes(self, client: TestClient) -> None:
+        """Test rows have data-sort attributes for client-side sorting."""
+        mock_containers = [
+            ContainerStats(
+                name="alpha",
+                host="nas",
+                status="running",
+                image="nginx:latest",
+                cpu_percent=10.0,
+                memory_usage=100,
+                memory_limit=1000,
+                memory_percent=10.0,
+                network_rx=100,
+                network_tx=100,
+                uptime="1 hour",
+                ports="",
+                engine="docker",
+                stack="alpha",
+                service="web",
+            ),
+        ]
+
+        with (
+            patch("compose_farm.web.routes.containers.get_config") as mock_config,
+            patch(
+                "compose_farm.web.routes.containers.fetch_all_container_stats",
+                new_callable=AsyncMock,
+            ) as mock_fetch,
+        ):
+            mock_config.return_value = Config(
+                compose_dir=Path("/opt/compose"),
+                hosts={"nas": Host(address="192.168.1.6")},
+                stacks={"test": "nas"},
+                glances_stack="glances",
+            )
+            mock_fetch.return_value = mock_containers
+
+            response = client.get("/api/containers/rows")
+            assert response.status_code == 200
+            # Check that cells have data-sort attributes
+            assert 'data-sort="alpha"' in response.text  # stack
+            assert 'data-sort="web"' in response.text  # service
+            assert 'data-sort="3600"' in response.text  # uptime (1 hour = 3600s)
+            assert 'data-sort="10' in response.text  # cpu

tests/test_executor.py

@@ -11,6 +11,7 @@ from compose_farm.executor import (
     _run_local_command,
     check_networks_exist,
     check_paths_exist,
+    get_running_stacks_on_host,
     is_local,
     run_command,
     run_compose,
@@ -239,3 +240,31 @@ class TestCheckNetworksExist:
 
         result = await check_networks_exist(config, "local", [])
         assert result == {}
+
+
+@linux_only
+class TestGetRunningStacksOnHost:
+    """Tests for get_running_stacks_on_host function (requires Docker)."""
+
+    async def test_returns_set_of_stacks(self, tmp_path: Path) -> None:
+        """Function returns a set of stack names."""
+        config = Config(
+            compose_dir=tmp_path,
+            hosts={"local": Host(address="localhost")},
+            stacks={},
+        )
+
+        result = await get_running_stacks_on_host(config, "local")
+        assert isinstance(result, set)
+
+    async def test_filters_empty_lines(self, tmp_path: Path) -> None:
+        """Empty project names are filtered out."""
+        config = Config(
+            compose_dir=tmp_path,
+            hosts={"local": Host(address="localhost")},
+            stacks={},
+        )
+
+        # Result should not contain empty strings
+        result = await get_running_stacks_on_host(config, "local")
+        assert "" not in result

tests/test_glances.py

@@ -0,0 +1,349 @@
+"""Tests for Glances integration."""
+
+from pathlib import Path
+from unittest.mock import AsyncMock, patch
+
+import httpx
+import pytest
+
+from compose_farm.config import Config, Host
+from compose_farm.glances import (
+    DEFAULT_GLANCES_PORT,
+    ContainerStats,
+    HostStats,
+    fetch_all_container_stats,
+    fetch_all_host_stats,
+    fetch_container_stats,
+    fetch_host_stats,
+)
+
+
+class TestHostStats:
+    """Tests for HostStats dataclass."""
+
+    def test_host_stats_creation(self) -> None:
+        stats = HostStats(
+            host="nas",
+            cpu_percent=25.5,
+            mem_percent=50.0,
+            swap_percent=10.0,
+            load=2.5,
+            disk_percent=75.0,
+        )
+        assert stats.host == "nas"
+        assert stats.cpu_percent == 25.5
+        assert stats.mem_percent == 50.0
+        assert stats.disk_percent == 75.0
+        assert stats.error is None
+
+    def test_host_stats_from_error(self) -> None:
+        stats = HostStats.from_error("nas", "Connection refused")
+        assert stats.host == "nas"
+        assert stats.cpu_percent == 0
+        assert stats.mem_percent == 0
+        assert stats.error == "Connection refused"
+
+
+class TestFetchHostStats:
+    """Tests for fetch_host_stats function."""
+
+    @pytest.mark.asyncio
+    async def test_fetch_host_stats_success(self) -> None:
+        quicklook_response = httpx.Response(
+            200,
+            json={
+                "cpu": 25.5,
+                "mem": 50.0,
+                "swap": 5.0,
+                "load": 2.5,
+            },
+        )
+        fs_response = httpx.Response(
+            200,
+            json=[
+                {"mnt_point": "/", "percent": 65.0},
+                {"mnt_point": "/mnt/data", "percent": 80.0},
+            ],
+        )
+
+        async def mock_get(url: str) -> httpx.Response:
+            if "quicklook" in url:
+                return quicklook_response
+            return fs_response
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(side_effect=mock_get)
+
+            stats = await fetch_host_stats("nas", "192.168.1.6")
+
+        assert stats.host == "nas"
+        assert stats.cpu_percent == 25.5
+        assert stats.mem_percent == 50.0
+        assert stats.swap_percent == 5.0
+        assert stats.load == 2.5
+        assert stats.disk_percent == 65.0  # Root filesystem
+        assert stats.error is None
+
+    @pytest.mark.asyncio
+    async def test_fetch_host_stats_http_error(self) -> None:
+        mock_response = httpx.Response(500)
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(return_value=mock_response)
+
+            stats = await fetch_host_stats("nas", "192.168.1.6")
+
+        assert stats.host == "nas"
+        assert stats.error == "HTTP 500"
+        assert stats.cpu_percent == 0
+
+    @pytest.mark.asyncio
+    async def test_fetch_host_stats_timeout(self) -> None:
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(side_effect=httpx.TimeoutException("timeout"))
+
+            stats = await fetch_host_stats("nas", "192.168.1.6")
+
+        assert stats.host == "nas"
+        assert stats.error == "timeout"
+
+    @pytest.mark.asyncio
+    async def test_fetch_host_stats_connection_error(self) -> None:
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(
+                side_effect=httpx.ConnectError("Connection refused")
+            )
+
+            stats = await fetch_host_stats("nas", "192.168.1.6")
+
+        assert stats.host == "nas"
+        assert stats.error is not None
+        assert "Connection refused" in stats.error
+
+
+class TestFetchAllHostStats:
+    """Tests for fetch_all_host_stats function."""
+
+    @pytest.mark.asyncio
+    async def test_fetch_all_host_stats(self) -> None:
+        config = Config(
+            compose_dir=Path("/opt/compose"),
+            hosts={
+                "nas": Host(address="192.168.1.6"),
+                "nuc": Host(address="192.168.1.2"),
+            },
+            stacks={"test": "nas"},
+        )
+
+        quicklook_response = httpx.Response(
+            200,
+            json={
+                "cpu": 25.5,
+                "mem": 50.0,
+                "swap": 5.0,
+                "load": 2.5,
+            },
+        )
+        fs_response = httpx.Response(
+            200,
+            json=[{"mnt_point": "/", "percent": 70.0}],
+        )
+
+        async def mock_get(url: str) -> httpx.Response:
+            if "quicklook" in url:
+                return quicklook_response
+            return fs_response
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(side_effect=mock_get)
+
+            stats = await fetch_all_host_stats(config)
+
+        assert "nas" in stats
+        assert "nuc" in stats
+        assert stats["nas"].cpu_percent == 25.5
+        assert stats["nuc"].cpu_percent == 25.5
+        assert stats["nas"].disk_percent == 70.0
+
+
+class TestDefaultPort:
+    """Tests for default Glances port constant."""
+
+    def test_default_port(self) -> None:
+        assert DEFAULT_GLANCES_PORT == 61208
+
+
+class TestContainerStats:
+    """Tests for ContainerStats dataclass."""
+
+    def test_container_stats_creation(self) -> None:
+        stats = ContainerStats(
+            name="nginx",
+            host="nas",
+            status="running",
+            image="nginx:latest",
+            cpu_percent=5.5,
+            memory_usage=104857600,  # 100MB
+            memory_limit=1073741824,  # 1GB
+            memory_percent=9.77,
+            network_rx=1000000,
+            network_tx=500000,
+            uptime="2 hours",
+            ports="80->80/tcp",
+            engine="docker",
+        )
+        assert stats.name == "nginx"
+        assert stats.host == "nas"
+        assert stats.cpu_percent == 5.5
+
+
+class TestFetchContainerStats:
+    """Tests for fetch_container_stats function."""
+
+    @pytest.mark.asyncio
+    async def test_fetch_container_stats_success(self) -> None:
+        mock_response = httpx.Response(
+            200,
+            json=[
+                {
+                    "name": "nginx",
+                    "status": "running",
+                    "image": ["nginx:latest"],
+                    "cpu_percent": 5.5,
+                    "memory_usage": 104857600,
+                    "memory_limit": 1073741824,
+                    "network": {"cumulative_rx": 1000, "cumulative_tx": 500},
+                    "uptime": "2 hours",
+                    "ports": "80->80/tcp",
+                    "engine": "docker",
+                },
+                {
+                    "name": "redis",
+                    "status": "running",
+                    "image": ["redis:7"],
+                    "cpu_percent": 1.2,
+                    "memory_usage": 52428800,
+                    "memory_limit": 1073741824,
+                    "network": {},
+                    "uptime": "3 hours",
+                    "ports": "",
+                    "engine": "docker",
+                },
+            ],
+        )
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(return_value=mock_response)
+
+            containers, error = await fetch_container_stats("nas", "192.168.1.6")
+
+        assert error is None
+        assert containers is not None
+        assert len(containers) == 2
+        assert containers[0].name == "nginx"
+        assert containers[0].host == "nas"
+        assert containers[0].cpu_percent == 5.5
+        assert containers[1].name == "redis"
+
+    @pytest.mark.asyncio
+    async def test_fetch_container_stats_empty_on_error(self) -> None:
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(side_effect=httpx.TimeoutException("timeout"))
+
+            containers, error = await fetch_container_stats("nas", "192.168.1.6")
+
+        assert containers is None
+        assert error == "Connection timed out"
+
+    @pytest.mark.asyncio
+    async def test_fetch_container_stats_handles_string_image(self) -> None:
+        """Test that image field works as string (not just list)."""
+        mock_response = httpx.Response(
+            200,
+            json=[
+                {
+                    "name": "test",
+                    "status": "running",
+                    "image": "myimage:v1",  # String instead of list
+                    "cpu_percent": 0,
+                    "memory_usage": 0,
+                    "memory_limit": 1,
+                    "network": {},
+                    "uptime": "",
+                    "ports": "",
+                    "engine": "docker",
+                },
+            ],
+        )
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(return_value=mock_response)
+
+            containers, error = await fetch_container_stats("nas", "192.168.1.6")
+
+        assert error is None
+        assert containers is not None
+        assert len(containers) == 1
+        assert containers[0].image == "myimage:v1"
+
+
+class TestFetchAllContainerStats:
+    """Tests for fetch_all_container_stats function."""
+
+    @pytest.mark.asyncio
+    async def test_fetch_all_container_stats(self) -> None:
+        config = Config(
+            compose_dir=Path("/opt/compose"),
+            hosts={
+                "nas": Host(address="192.168.1.6"),
+                "nuc": Host(address="192.168.1.2"),
+            },
+            stacks={"test": "nas"},
+        )
+
+        mock_response = httpx.Response(
+            200,
+            json=[
+                {
+                    "name": "nginx",
+                    "status": "running",
+                    "image": ["nginx:latest"],
+                    "cpu_percent": 5.5,
+                    "memory_usage": 104857600,
+                    "memory_limit": 1073741824,
+                    "network": {},
+                    "uptime": "2 hours",
+                    "ports": "",
+                    "engine": "docker",
+                },
+            ],
+        )
+
+        with patch("httpx.AsyncClient") as mock_client:
+            mock_client.return_value.__aenter__ = AsyncMock(return_value=mock_client.return_value)
+            mock_client.return_value.__aexit__ = AsyncMock(return_value=None)
+            mock_client.return_value.get = AsyncMock(return_value=mock_response)
+
+            containers = await fetch_all_container_stats(config)
+
+        # 2 hosts x 1 container each = 2 containers
+        assert len(containers) == 2
+        hosts = {c.host for c in containers}
+        assert "nas" in hosts
+        assert "nuc" in hosts

tests/test_logs.py

@@ -10,8 +10,8 @@ import pytest
 from compose_farm.config import Config, Host
 from compose_farm.executor import CommandResult
 from compose_farm.logs import (
-    _parse_images_output,
-    collect_stack_entries,
+    _SECTION_SEPARATOR,
+    collect_stacks_entries_on_host,
     isoformat,
     load_existing_entries,
     merge_entries,
@@ -19,74 +19,252 @@ from compose_farm.logs import (
 )
 
 
-def test_parse_images_output_handles_list_and_lines() -> None:
-    data = [
-        {"Service": "svc", "Image": "redis", "Digest": "sha256:abc"},
-        {"Service": "svc", "Image": "db", "Digest": "sha256:def"},
+def _make_mock_output(
+    project_images: dict[str, list[str]], image_info: list[dict[str, object]]
+) -> str:
+    """Build mock output matching the 2-docker-command format."""
+    # Section 1: project|image pairs from docker ps
+    ps_lines = [
+        f"{project}|{image}" for project, images in project_images.items() for image in images
     ]
-    as_array = _parse_images_output(json.dumps(data))
-    assert len(as_array) == 2
 
-    as_lines = _parse_images_output("\n".join(json.dumps(item) for item in data))
-    assert len(as_lines) == 2
+    # Section 2: JSON array from docker image inspect
+    image_json = json.dumps(image_info)
+
+    return f"{chr(10).join(ps_lines)}\n{_SECTION_SEPARATOR}\n{image_json}"
 
 
-@pytest.mark.asyncio
-async def test_snapshot_preserves_first_seen(tmp_path: Path) -> None:
-    compose_dir = tmp_path / "compose"
-    compose_dir.mkdir()
-    stack_dir = compose_dir / "svc"
-    stack_dir.mkdir()
-    (stack_dir / "docker-compose.yml").write_text("services: {}\n")
+class TestCollectStacksEntriesOnHost:
+    """Tests for collect_stacks_entries_on_host (2 docker commands per host)."""
 
-    config = Config(
-        compose_dir=compose_dir,
-        hosts={"local": Host(address="localhost")},
-        stacks={"svc": "local"},
-    )
+    @pytest.fixture
+    def config_with_stacks(self, tmp_path: Path) -> Config:
+        """Create a config with multiple stacks."""
+        compose_dir = tmp_path / "compose"
+        compose_dir.mkdir()
+        for stack in ["plex", "jellyfin", "sonarr"]:
+            stack_dir = compose_dir / stack
+            stack_dir.mkdir()
+            (stack_dir / "docker-compose.yml").write_text("services: {}\n")
 
-    sample_output = json.dumps([{"Service": "svc", "Image": "redis", "Digest": "sha256:abc"}])
-
-    async def fake_run_compose(
-        _cfg: Config, stack: str, compose_cmd: str, *, stream: bool = True
-    ) -> CommandResult:
-        assert compose_cmd == "images --format json"
-        assert stream is False or stream is True
-        return CommandResult(
-            stack=stack,
-            exit_code=0,
-            success=True,
-            stdout=sample_output,
-            stderr="",
+        return Config(
+            compose_dir=compose_dir,
+            hosts={"host1": Host(address="localhost"), "host2": Host(address="localhost")},
+            stacks={"plex": "host1", "jellyfin": "host1", "sonarr": "host2"},
         )
 
-    log_path = tmp_path / "dockerfarm-log.toml"
+    @pytest.mark.asyncio
+    async def test_single_ssh_call(
+        self, config_with_stacks: Config, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """Verify only 1 SSH call is made regardless of stack count."""
+        call_count = {"count": 0}
 
-    # First snapshot
-    first_time = datetime(2025, 1, 1, tzinfo=UTC)
-    first_entries = await collect_stack_entries(
-        config, "svc", now=first_time, run_compose_fn=fake_run_compose
-    )
-    first_iso = isoformat(first_time)
-    merged = merge_entries([], first_entries, now_iso=first_iso)
-    meta = {"generated_at": first_iso, "compose_dir": str(config.compose_dir)}
-    write_toml(log_path, meta=meta, entries=merged)
+        async def mock_run_command(
+            host: Host, command: str, stack: str, *, stream: bool, prefix: str
+        ) -> CommandResult:
+            call_count["count"] += 1
+            output = _make_mock_output(
+                {"plex": ["plex:latest"], "jellyfin": ["jellyfin:latest"]},
+                [
+                    {
+                        "RepoTags": ["plex:latest"],
+                        "Id": "sha256:aaa",
+                        "RepoDigests": ["plex@sha256:aaa"],
+                    },
+                    {
+                        "RepoTags": ["jellyfin:latest"],
+                        "Id": "sha256:bbb",
+                        "RepoDigests": ["jellyfin@sha256:bbb"],
+                    },
+                ],
+            )
+            return CommandResult(stack=stack, exit_code=0, success=True, stdout=output)
 
-    after_first = tomllib.loads(log_path.read_text())
-    first_seen = after_first["entries"][0]["first_seen"]
+        monkeypatch.setattr("compose_farm.logs.run_command", mock_run_command)
 
-    # Second snapshot
-    second_time = datetime(2025, 2, 1, tzinfo=UTC)
-    second_entries = await collect_stack_entries(
-        config, "svc", now=second_time, run_compose_fn=fake_run_compose
-    )
-    second_iso = isoformat(second_time)
-    existing = load_existing_entries(log_path)
-    merged = merge_entries(existing, second_entries, now_iso=second_iso)
-    meta = {"generated_at": second_iso, "compose_dir": str(config.compose_dir)}
-    write_toml(log_path, meta=meta, entries=merged)
+        now = datetime(2025, 1, 1, tzinfo=UTC)
+        entries = await collect_stacks_entries_on_host(
+            config_with_stacks, "host1", {"plex", "jellyfin"}, now=now
+        )
 
-    after_second = tomllib.loads(log_path.read_text())
-    entry = after_second["entries"][0]
-    assert entry["first_seen"] == first_seen
-    assert entry["last_seen"].startswith("2025-02-01")
+        assert call_count["count"] == 1
+        assert len(entries) == 2
+
+    @pytest.mark.asyncio
+    async def test_filters_to_requested_stacks(
+        self, config_with_stacks: Config, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """Only return entries for stacks we asked for, even if others are running."""
+
+        async def mock_run_command(
+            host: Host, command: str, stack: str, *, stream: bool, prefix: str
+        ) -> CommandResult:
+            # Docker ps shows 3 stacks, but we only want plex
+            output = _make_mock_output(
+                {
+                    "plex": ["plex:latest"],
+                    "jellyfin": ["jellyfin:latest"],
+                    "other": ["other:latest"],
+                },
+                [
+                    {
+                        "RepoTags": ["plex:latest"],
+                        "Id": "sha256:aaa",
+                        "RepoDigests": ["plex@sha256:aaa"],
+                    },
+                    {
+                        "RepoTags": ["jellyfin:latest"],
+                        "Id": "sha256:bbb",
+                        "RepoDigests": ["j@sha256:bbb"],
+                    },
+                    {
+                        "RepoTags": ["other:latest"],
+                        "Id": "sha256:ccc",
+                        "RepoDigests": ["o@sha256:ccc"],
+                    },
+                ],
+            )
+            return CommandResult(stack=stack, exit_code=0, success=True, stdout=output)
+
+        monkeypatch.setattr("compose_farm.logs.run_command", mock_run_command)
+
+        now = datetime(2025, 1, 1, tzinfo=UTC)
+        entries = await collect_stacks_entries_on_host(
+            config_with_stacks, "host1", {"plex"}, now=now
+        )
+
+        assert len(entries) == 1
+        assert entries[0].stack == "plex"
+
+    @pytest.mark.asyncio
+    async def test_multiple_images_per_stack(
+        self, config_with_stacks: Config, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """Stack with multiple containers/images returns multiple entries."""
+
+        async def mock_run_command(
+            host: Host, command: str, stack: str, *, stream: bool, prefix: str
+        ) -> CommandResult:
+            output = _make_mock_output(
+                {"plex": ["plex:latest", "redis:7"]},
+                [
+                    {
+                        "RepoTags": ["plex:latest"],
+                        "Id": "sha256:aaa",
+                        "RepoDigests": ["p@sha256:aaa"],
+                    },
+                    {"RepoTags": ["redis:7"], "Id": "sha256:bbb", "RepoDigests": ["r@sha256:bbb"]},
+                ],
+            )
+            return CommandResult(stack=stack, exit_code=0, success=True, stdout=output)
+
+        monkeypatch.setattr("compose_farm.logs.run_command", mock_run_command)
+
+        now = datetime(2025, 1, 1, tzinfo=UTC)
+        entries = await collect_stacks_entries_on_host(
+            config_with_stacks, "host1", {"plex"}, now=now
+        )
+
+        assert len(entries) == 2
+        images = {e.image for e in entries}
+        assert images == {"plex:latest", "redis:7"}
+
+    @pytest.mark.asyncio
+    async def test_empty_stacks_returns_empty(self, config_with_stacks: Config) -> None:
+        """Empty stack set returns empty entries without making SSH call."""
+        now = datetime(2025, 1, 1, tzinfo=UTC)
+        entries = await collect_stacks_entries_on_host(config_with_stacks, "host1", set(), now=now)
+        assert entries == []
+
+    @pytest.mark.asyncio
+    async def test_ssh_failure_returns_empty(
+        self, config_with_stacks: Config, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """SSH failure returns empty list instead of raising."""
+
+        async def mock_run_command(
+            host: Host, command: str, stack: str, *, stream: bool, prefix: str
+        ) -> CommandResult:
+            return CommandResult(stack=stack, exit_code=1, success=False, stdout="", stderr="error")
+
+        monkeypatch.setattr("compose_farm.logs.run_command", mock_run_command)
+
+        now = datetime(2025, 1, 1, tzinfo=UTC)
+        entries = await collect_stacks_entries_on_host(
+            config_with_stacks, "host1", {"plex"}, now=now
+        )
+
+        assert entries == []
+
+
+class TestSnapshotMerging:
+    """Tests for merge_entries preserving first_seen."""
+
+    @pytest.fixture
+    def config(self, tmp_path: Path) -> Config:
+        compose_dir = tmp_path / "compose"
+        compose_dir.mkdir()
+        stack_dir = compose_dir / "svc"
+        stack_dir.mkdir()
+        (stack_dir / "docker-compose.yml").write_text("services: {}\n")
+
+        return Config(
+            compose_dir=compose_dir,
+            hosts={"local": Host(address="localhost")},
+            stacks={"svc": "local"},
+        )
+
+    @pytest.mark.asyncio
+    async def test_preserves_first_seen(
+        self, tmp_path: Path, config: Config, monkeypatch: pytest.MonkeyPatch
+    ) -> None:
+        """Repeated snapshots preserve first_seen timestamp."""
+
+        async def mock_run_command(
+            host: Host, command: str, stack: str, *, stream: bool, prefix: str
+        ) -> CommandResult:
+            output = _make_mock_output(
+                {"svc": ["redis:latest"]},
+                [
+                    {
+                        "RepoTags": ["redis:latest"],
+                        "Id": "sha256:abc",
+                        "RepoDigests": ["r@sha256:abc"],
+                    }
+                ],
+            )
+            return CommandResult(stack=stack, exit_code=0, success=True, stdout=output)
+
+        monkeypatch.setattr("compose_farm.logs.run_command", mock_run_command)
+
+        log_path = tmp_path / "dockerfarm-log.toml"
+
+        # First snapshot
+        first_time = datetime(2025, 1, 1, tzinfo=UTC)
+        first_entries = await collect_stacks_entries_on_host(
+            config, "local", {"svc"}, now=first_time
+        )
+        first_iso = isoformat(first_time)
+        merged = merge_entries([], first_entries, now_iso=first_iso)
+        meta = {"generated_at": first_iso, "compose_dir": str(config.compose_dir)}
+        write_toml(log_path, meta=meta, entries=merged)
+
+        after_first = tomllib.loads(log_path.read_text())
+        first_seen = after_first["entries"][0]["first_seen"]
+
+        # Second snapshot
+        second_time = datetime(2025, 2, 1, tzinfo=UTC)
+        second_entries = await collect_stacks_entries_on_host(
+            config, "local", {"svc"}, now=second_time
+        )
+        second_iso = isoformat(second_time)
+        existing = load_existing_entries(log_path)
+        merged = merge_entries(existing, second_entries, now_iso=second_iso)
+        meta = {"generated_at": second_iso, "compose_dir": str(config.compose_dir)}
+        write_toml(log_path, meta=meta, entries=merged)
+
+        after_second = tomllib.loads(log_path.read_text())
+        entry = after_second["entries"][0]
+        assert entry["first_seen"] == first_seen
+        assert entry["last_seen"].startswith("2025-02-01")

tests/test_operations.py

@@ -11,7 +11,10 @@ import pytest
 from compose_farm.cli import lifecycle
 from compose_farm.config import Config, Host
 from compose_farm.executor import CommandResult
-from compose_farm.operations import _migrate_stack
+from compose_farm.operations import (
+    _migrate_stack,
+    build_discovery_results,
+)
 
 
 @pytest.fixture
@@ -109,3 +112,83 @@ class TestUpdateCommandSequence:
         # Verify the sequence is pull, build, down, up
         assert "down" in source
         assert "up -d" in source
+
+
+class TestBuildDiscoveryResults:
+    """Tests for build_discovery_results function."""
+
+    @pytest.fixture
+    def config(self, tmp_path: Path) -> Config:
+        """Create a test config with multiple stacks."""
+        compose_dir = tmp_path / "compose"
+        for stack in ["plex", "jellyfin", "sonarr"]:
+            (compose_dir / stack).mkdir(parents=True)
+            (compose_dir / stack / "docker-compose.yml").write_text("services: {}")
+
+        return Config(
+            compose_dir=compose_dir,
+            hosts={
+                "host1": Host(address="localhost"),
+                "host2": Host(address="localhost"),
+            },
+            stacks={"plex": "host1", "jellyfin": "host1", "sonarr": "host2"},
+        )
+
+    def test_discovers_correctly_running_stacks(self, config: Config) -> None:
+        """Stacks running on correct hosts are discovered."""
+        running_on_host = {
+            "host1": {"plex", "jellyfin"},
+            "host2": {"sonarr"},
+        }
+
+        discovered, strays, duplicates = build_discovery_results(config, running_on_host)
+
+        assert discovered == {"plex": "host1", "jellyfin": "host1", "sonarr": "host2"}
+        assert strays == {}
+        assert duplicates == {}
+
+    def test_detects_stray_stacks(self, config: Config) -> None:
+        """Stacks running on wrong hosts are marked as strays."""
+        running_on_host = {
+            "host1": set(),
+            "host2": {"plex"},  # plex should be on host1
+        }
+
+        discovered, strays, _duplicates = build_discovery_results(config, running_on_host)
+
+        assert "plex" not in discovered
+        assert strays == {"plex": ["host2"]}
+
+    def test_detects_duplicates(self, config: Config) -> None:
+        """Single-host stacks running on multiple hosts are duplicates."""
+        running_on_host = {
+            "host1": {"plex"},
+            "host2": {"plex"},  # plex running on both hosts
+        }
+
+        discovered, strays, duplicates = build_discovery_results(
+            config, running_on_host, stacks=["plex"]
+        )
+
+        # plex is correctly running on host1
+        assert discovered == {"plex": "host1"}
+        # plex is also a stray on host2
+        assert strays == {"plex": ["host2"]}
+        # plex is a duplicate (single-host stack on multiple hosts)
+        assert duplicates == {"plex": ["host1", "host2"]}
+
+    def test_filters_to_requested_stacks(self, config: Config) -> None:
+        """Only returns results for requested stacks."""
+        running_on_host = {
+            "host1": {"plex", "jellyfin"},
+            "host2": {"sonarr"},
+        }
+
+        discovered, _strays, _duplicates = build_discovery_results(
+            config, running_on_host, stacks=["plex"]
+        )
+
+        # Only plex should be in results
+        assert discovered == {"plex": "host1"}
+        assert "jellyfin" not in discovered
+        assert "sonarr" not in discovered

tests/test_refresh.py

@@ -211,8 +211,8 @@ class TestRefreshCommand:
                 return_value=existing_state,
             ),
             patch(
-                "compose_farm.cli.management._discover_stacks",
-                return_value={"plex": "nas02"},  # plex moved to nas02
+                "compose_farm.cli.management._discover_stacks_full",
+                return_value=({"plex": "nas02"}, {}, {}),  # plex moved to nas02
             ),
             patch("compose_farm.cli.management._snapshot_stacks"),
             patch("compose_farm.cli.management.save_state") as mock_save,
@@ -247,8 +247,12 @@ class TestRefreshCommand:
                 return_value=existing_state,
             ),
             patch(
-                "compose_farm.cli.management._discover_stacks",
-                return_value={"plex": "nas01", "grafana": "nas02"},  # jellyfin not running
+                "compose_farm.cli.management._discover_stacks_full",
+                return_value=(
+                    {"plex": "nas01", "grafana": "nas02"},
+                    {},
+                    {},
+                ),  # jellyfin not running
             ),
             patch("compose_farm.cli.management._snapshot_stacks"),
             patch("compose_farm.cli.management.save_state") as mock_save,
@@ -281,8 +285,8 @@ class TestRefreshCommand:
                 return_value=existing_state,
             ),
             patch(
-                "compose_farm.cli.management._discover_stacks",
-                return_value={"plex": "nas01"},  # only plex running
+                "compose_farm.cli.management._discover_stacks_full",
+                return_value=({"plex": "nas01"}, {}, {}),  # only plex running
             ),
             patch("compose_farm.cli.management._snapshot_stacks"),
             patch("compose_farm.cli.management.save_state") as mock_save,
@@ -315,8 +319,8 @@ class TestRefreshCommand:
                 return_value=existing_state,
             ),
             patch(
-                "compose_farm.cli.management._discover_stacks",
-                return_value={"plex": "nas01"},  # jellyfin not running
+                "compose_farm.cli.management._discover_stacks_full",
+                return_value=({"plex": "nas01"}, {}, {}),  # jellyfin not running
             ),
             patch("compose_farm.cli.management._snapshot_stacks"),
             patch("compose_farm.cli.management.save_state") as mock_save,
@@ -350,8 +354,8 @@ class TestRefreshCommand:
                 return_value=existing_state,
             ),
             patch(
-                "compose_farm.cli.management._discover_stacks",
-                return_value={"plex": "nas02"},  # would change
+                "compose_farm.cli.management._discover_stacks_full",
+                return_value=({"plex": "nas02"}, {}, {}),  # would change
             ),
             patch("compose_farm.cli.management.save_state") as mock_save,
         ):

tests/test_registry.py

@@ -0,0 +1,182 @@
+"""Tests for registry module."""
+
+from compose_farm.registry import (
+    DOCKER_HUB_ALIASES,
+    ImageRef,
+    RegistryClient,
+    TagCheckResult,
+    _find_updates,
+    _parse_version,
+)
+
+
+class TestImageRef:
+    """Tests for ImageRef parsing."""
+
+    def test_parse_simple_image(self) -> None:
+        """Test parsing simple image name."""
+        ref = ImageRef.parse("nginx")
+        assert ref.registry == "docker.io"
+        assert ref.namespace == "library"
+        assert ref.name == "nginx"
+        assert ref.tag == "latest"
+
+    def test_parse_image_with_tag(self) -> None:
+        """Test parsing image with tag."""
+        ref = ImageRef.parse("nginx:1.25")
+        assert ref.registry == "docker.io"
+        assert ref.namespace == "library"
+        assert ref.name == "nginx"
+        assert ref.tag == "1.25"
+
+    def test_parse_image_with_namespace(self) -> None:
+        """Test parsing image with namespace."""
+        ref = ImageRef.parse("linuxserver/jellyfin:latest")
+        assert ref.registry == "docker.io"
+        assert ref.namespace == "linuxserver"
+        assert ref.name == "jellyfin"
+        assert ref.tag == "latest"
+
+    def test_parse_ghcr_image(self) -> None:
+        """Test parsing GitHub Container Registry image."""
+        ref = ImageRef.parse("ghcr.io/user/repo:v1.0.0")
+        assert ref.registry == "ghcr.io"
+        assert ref.namespace == "user"
+        assert ref.name == "repo"
+        assert ref.tag == "v1.0.0"
+
+    def test_parse_image_with_digest(self) -> None:
+        """Test parsing image with digest."""
+        ref = ImageRef.parse("nginx:latest@sha256:abc123")
+        assert ref.registry == "docker.io"
+        assert ref.name == "nginx"
+        assert ref.tag == "latest"
+        assert ref.digest == "sha256:abc123"
+
+    def test_full_name_with_namespace(self) -> None:
+        """Test full_name property with namespace."""
+        ref = ImageRef.parse("linuxserver/jellyfin")
+        assert ref.full_name == "linuxserver/jellyfin"
+
+    def test_full_name_without_namespace(self) -> None:
+        """Test full_name property for official images."""
+        ref = ImageRef.parse("nginx")
+        assert ref.full_name == "library/nginx"
+
+    def test_display_name_official_image(self) -> None:
+        """Test display_name for official Docker Hub images."""
+        ref = ImageRef.parse("nginx:latest")
+        assert ref.display_name == "nginx"
+
+    def test_display_name_hub_with_namespace(self) -> None:
+        """Test display_name for Docker Hub images with namespace."""
+        ref = ImageRef.parse("linuxserver/jellyfin")
+        assert ref.display_name == "linuxserver/jellyfin"
+
+    def test_display_name_other_registry(self) -> None:
+        """Test display_name for other registries."""
+        ref = ImageRef.parse("ghcr.io/user/repo")
+        assert ref.display_name == "ghcr.io/user/repo"
+
+
+class TestParseVersion:
+    """Tests for version parsing."""
+
+    def test_parse_semver(self) -> None:
+        """Test parsing semantic version."""
+        assert _parse_version("1.2.3") == (1, 2, 3)
+
+    def test_parse_version_with_v_prefix(self) -> None:
+        """Test parsing version with v prefix."""
+        assert _parse_version("v1.2.3") == (1, 2, 3)
+        assert _parse_version("V1.2.3") == (1, 2, 3)
+
+    def test_parse_two_part_version(self) -> None:
+        """Test parsing two-part version."""
+        assert _parse_version("1.25") == (1, 25)
+
+    def test_parse_single_number(self) -> None:
+        """Test parsing single number version."""
+        assert _parse_version("7") == (7,)
+
+    def test_parse_invalid_version(self) -> None:
+        """Test parsing non-version tags."""
+        assert _parse_version("latest") is None
+        assert _parse_version("stable") is None
+        assert _parse_version("alpine") is None
+
+
+class TestFindUpdates:
+    """Tests for finding available updates."""
+
+    def test_find_updates_with_newer_versions(self) -> None:
+        """Test finding newer versions."""
+        current = "1.0.0"
+        tags = ["0.9.0", "1.0.0", "1.1.0", "2.0.0"]
+        updates = _find_updates(current, tags)
+        assert updates == ["2.0.0", "1.1.0"]
+
+    def test_find_updates_no_newer(self) -> None:
+        """Test when already on latest."""
+        current = "2.0.0"
+        tags = ["1.0.0", "1.5.0", "2.0.0"]
+        updates = _find_updates(current, tags)
+        assert updates == []
+
+    def test_find_updates_non_version_tag(self) -> None:
+        """Test with non-version current tag."""
+        current = "latest"
+        tags = ["1.0.0", "2.0.0"]
+        updates = _find_updates(current, tags)
+        # Can't determine updates for non-version tags
+        assert updates == []
+
+
+class TestRegistryClient:
+    """Tests for unified registry client."""
+
+    def test_docker_hub_normalization(self) -> None:
+        """Test Docker Hub aliases are normalized."""
+        for alias in DOCKER_HUB_ALIASES:
+            client = RegistryClient(alias)
+            assert client.registry == "docker.io"
+            assert client.registry_url == "https://registry-1.docker.io"
+
+    def test_ghcr_client(self) -> None:
+        """Test GitHub Container Registry client."""
+        client = RegistryClient("ghcr.io")
+        assert client.registry == "ghcr.io"
+        assert client.registry_url == "https://ghcr.io"
+
+    def test_generic_registry(self) -> None:
+        """Test generic registry client."""
+        client = RegistryClient("quay.io")
+        assert client.registry == "quay.io"
+        assert client.registry_url == "https://quay.io"
+
+
+class TestTagCheckResult:
+    """Tests for TagCheckResult."""
+
+    def test_create_result(self) -> None:
+        """Test creating a result."""
+        ref = ImageRef.parse("nginx:1.25")
+        result = TagCheckResult(
+            image=ref,
+            current_digest="sha256:abc",
+            available_updates=["1.26", "1.27"],
+        )
+        assert result.image.name == "nginx"
+        assert result.available_updates == ["1.26", "1.27"]
+        assert result.error is None
+
+    def test_result_with_error(self) -> None:
+        """Test result with error."""
+        ref = ImageRef.parse("nginx")
+        result = TagCheckResult(
+            image=ref,
+            current_digest="",
+            error="Connection refused",
+        )
+        assert result.error == "Connection refused"
+        assert result.available_updates == []

tests/test_ssh_keys.py

@@ -219,7 +219,7 @@ class TestSshConnectKwargs:
             assert result["client_keys"] == [str(key_path)]
 
     def test_includes_both_agent_and_key(self, tmp_path: Path) -> None:
-        """Include both agent_path and client_keys when both available."""
+        """Prioritize client_keys over agent_path when both available."""
         host = Host(address="example.com")
         key_path = tmp_path / "compose-farm"
 
@@ -229,7 +229,8 @@ class TestSshConnectKwargs:
         ):
             result = ssh_connect_kwargs(host)
 
-            assert result["agent_path"] == "/tmp/agent.sock"
+            # Agent should be ignored in favor of the dedicated key
+            assert "agent_path" not in result
             assert result["client_keys"] == [str(key_path)]
 
     def test_custom_port(self) -> None:

tests/web/test_backup.py

@@ -2,53 +2,65 @@
 
 from pathlib import Path
 
+import pytest
+
 from compose_farm.web.routes.api import _backup_file, _save_with_backup
 
 
-def test_backup_creates_timestamped_file(tmp_path: Path) -> None:
-    """Test that backup creates file in .backups with correct content."""
-    test_file = tmp_path / "test.yaml"
+@pytest.fixture
+def xdg_backup_dir(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> Path:
+    """Set XDG_CONFIG_HOME to tmp_path and return the backup directory path."""
+    monkeypatch.setenv("XDG_CONFIG_HOME", str(tmp_path))
+    return tmp_path / "compose-farm" / "backups"
+
+
+def test_backup_creates_timestamped_file(tmp_path: Path, xdg_backup_dir: Path) -> None:
+    """Test that backup creates file in XDG backup dir with correct content."""
+    test_file = tmp_path / "stacks" / "test.yaml"
+    test_file.parent.mkdir(parents=True)
     test_file.write_text("original content")
 
     backup_path = _backup_file(test_file)
 
     assert backup_path is not None
-    assert backup_path.parent.name == ".backups"
+    assert backup_path.is_relative_to(xdg_backup_dir)
     assert backup_path.name.startswith("test.yaml.")
     assert backup_path.read_text() == "original content"
 
 
-def test_backup_returns_none_for_nonexistent_file(tmp_path: Path) -> None:
+def test_backup_returns_none_for_nonexistent_file(tmp_path: Path, xdg_backup_dir: Path) -> None:
     """Test that backup returns None if file doesn't exist."""
     assert _backup_file(tmp_path / "nonexistent.yaml") is None
 
 
-def test_save_creates_new_file(tmp_path: Path) -> None:
+def test_save_creates_new_file(tmp_path: Path, xdg_backup_dir: Path) -> None:
     """Test that save creates new file without backup."""
     test_file = tmp_path / "new.yaml"
 
     assert _save_with_backup(test_file, "content") is True
     assert test_file.read_text() == "content"
-    assert not (tmp_path / ".backups").exists()
+    assert not xdg_backup_dir.exists()
 
 
-def test_save_skips_unchanged_content(tmp_path: Path) -> None:
+def test_save_skips_unchanged_content(tmp_path: Path, xdg_backup_dir: Path) -> None:
     """Test that save returns False and creates no backup if unchanged."""
     test_file = tmp_path / "test.yaml"
     test_file.write_text("same")
 
     assert _save_with_backup(test_file, "same") is False
-    assert not (tmp_path / ".backups").exists()
+    assert not xdg_backup_dir.exists()
 
 
-def test_save_creates_backup_before_overwrite(tmp_path: Path) -> None:
+def test_save_creates_backup_before_overwrite(tmp_path: Path, xdg_backup_dir: Path) -> None:
     """Test that save backs up original before overwriting."""
-    test_file = tmp_path / "test.yaml"
+    test_file = tmp_path / "stacks" / "test.yaml"
+    test_file.parent.mkdir(parents=True)
     test_file.write_text("original")
 
     assert _save_with_backup(test_file, "new") is True
     assert test_file.read_text() == "new"
 
-    backups = list((tmp_path / ".backups").glob("test.yaml.*"))
+    # Find backup in XDG dir
+    backups = list(xdg_backup_dir.rglob("test.yaml.*"))
     assert len(backups) == 1
     assert backups[0].read_text() == "original"

tests/web/test_helpers.py

@@ -7,11 +7,58 @@ from typing import TYPE_CHECKING
 
 import pytest
 from fastapi import HTTPException
+from pydantic import ValidationError
 
 if TYPE_CHECKING:
     from compose_farm.config import Config
 
 
+class TestExtractConfigError:
+    """Tests for extract_config_error helper."""
+
+    def test_validation_error_with_location(self) -> None:
+        from compose_farm.config import Config, Host
+        from compose_farm.web.deps import extract_config_error
+
+        # Trigger a validation error with an extra field
+        with pytest.raises(ValidationError) as exc_info:
+            Config(
+                hosts={"server": Host(address="192.168.1.1")},
+                stacks={"app": "server"},
+                unknown_field="bad",  # type: ignore[call-arg]
+            )
+
+        msg = extract_config_error(exc_info.value)
+        assert "unknown_field" in msg
+        assert "Extra inputs are not permitted" in msg
+
+    def test_validation_error_nested_location(self) -> None:
+        from compose_farm.config import Host
+        from compose_farm.web.deps import extract_config_error
+
+        # Trigger a validation error with a nested extra field
+        with pytest.raises(ValidationError) as exc_info:
+            Host(address="192.168.1.1", bad_key="value")  # type: ignore[call-arg]
+
+        msg = extract_config_error(exc_info.value)
+        assert "bad_key" in msg
+        assert "Extra inputs are not permitted" in msg
+
+    def test_regular_exception(self) -> None:
+        from compose_farm.web.deps import extract_config_error
+
+        exc = ValueError("Something went wrong")
+        msg = extract_config_error(exc)
+        assert msg == "Something went wrong"
+
+    def test_file_not_found_exception(self) -> None:
+        from compose_farm.web.deps import extract_config_error
+
+        exc = FileNotFoundError("Config file not found")
+        msg = extract_config_error(exc)
+        assert msg == "Config file not found"
+
+
 class TestValidateYaml:
     """Tests for _validate_yaml helper."""
 

tests/web/test_htmx_browser.py

@@ -134,6 +134,13 @@ def test_config(tmp_path_factory: pytest.TempPathFactory) -> Path:
         else:
             (svc / "compose.yaml").write_text(f"services:\n  {name}:\n    image: test/{name}\n")
 
+    # Create glances stack (required for containers page)
+    glances_dir = compose_dir / "glances"
+    glances_dir.mkdir()
+    (glances_dir / "compose.yaml").write_text(
+        "services:\n  glances:\n    image: nicolargo/glances\n"
+    )
+
     # Create config with multiple hosts
     config = tmp / "compose-farm.yaml"
     config.write_text(f"""
@@ -151,6 +158,8 @@ stacks:
   nextcloud: server-2
   jellyfin: server-2
   redis: server-1
+  glances: all
+glances_stack: glances
 """)
 
     # Create state (plex and nextcloud running, grafana and jellyfin not started)
@@ -245,7 +254,7 @@ class TestHTMXSidebarLoading:
 
         # Verify actual stacks from test config appear
         stacks = page.locator("#sidebar-stacks li")
-        assert stacks.count() == 5  # plex, grafana, nextcloud, jellyfin, redis
+        assert stacks.count() == 6  # plex, grafana, nextcloud, jellyfin, redis, glances
 
         # Check specific stacks are present
         content = page.locator("#sidebar-stacks").inner_text()
@@ -348,7 +357,7 @@ class TestDashboardContent:
 
         # From test config: 2 hosts, 5 stacks, 2 running (plex, nextcloud)
         assert "2" in stats  # hosts count
-        assert "5" in stats  # stacks count
+        assert "6" in stats  # stacks count
 
     def test_pending_shows_not_started_stacks(self, page: Page, server_url: str) -> None:
         """Pending operations shows grafana and jellyfin as not started."""
@@ -476,9 +485,9 @@ class TestSidebarFilter:
         page.goto(server_url)
         page.wait_for_selector("#sidebar-stacks", timeout=TIMEOUT)
 
-        # Initially all 4 stacks visible
+        # Initially all 6 stacks visible
         visible_items = page.locator("#sidebar-stacks li:not([hidden])")
-        assert visible_items.count() == 5
+        assert visible_items.count() == 6
 
         # Type in filter to match only "plex"
         self._filter_sidebar(page, "plex")
@@ -493,9 +502,9 @@ class TestSidebarFilter:
         page.goto(server_url)
         page.wait_for_selector("#sidebar-stacks", timeout=TIMEOUT)
 
-        # Initial count should be (5)
+        # Initial count should be (6)
         count_badge = page.locator("#sidebar-count")
-        assert "(5)" in count_badge.inner_text()
+        assert "(6)" in count_badge.inner_text()
 
         # Filter to show only stacks containing "x" (plex, nextcloud)
         self._filter_sidebar(page, "x")
@@ -524,13 +533,14 @@ class TestSidebarFilter:
         # Select server-1 from dropdown
         page.locator("#sidebar-host-select").select_option("server-1")
 
-        # Only plex, grafana, and redis (server-1 stacks) should be visible
+        # plex, grafana, redis (server-1), and glances (all) should be visible
         visible = page.locator("#sidebar-stacks li:not([hidden])")
-        assert visible.count() == 3
+        assert visible.count() == 4
 
         content = visible.all_inner_texts()
         assert any("plex" in s for s in content)
         assert any("grafana" in s for s in content)
+        assert any("glances" in s for s in content)
         assert not any("nextcloud" in s for s in content)
         assert not any("jellyfin" in s for s in content)
 
@@ -562,7 +572,7 @@ class TestSidebarFilter:
         self._filter_sidebar(page, "")
 
         # All stacks visible again
-        assert page.locator("#sidebar-stacks li:not([hidden])").count() == 5
+        assert page.locator("#sidebar-stacks li:not([hidden])").count() == 6
 
 
 class TestCommandPalette:
@@ -884,7 +894,7 @@ class TestContentStability:
 
         # Remember sidebar state
         initial_count = page.locator("#sidebar-stacks li").count()
-        assert initial_count == 5
+        assert initial_count == 6
 
         # Navigate away
         page.locator("#sidebar-stacks a", has_text="plex").click()
@@ -2329,3 +2339,227 @@ class TestTerminalNavigationIsolation:
         # Terminal should still be collapsed (no task to reconnect to)
         terminal_toggle = page.locator("#terminal-toggle")
         assert not terminal_toggle.is_checked(), "Terminal should remain collapsed after navigation"
+
+
+class TestContainersPagePause:
+    """Test containers page auto-refresh pause mechanism.
+
+    The containers page auto-refreshes every 3 seconds. When a user opens
+    an action dropdown, refresh should pause to prevent the dropdown from
+    closing unexpectedly.
+    """
+
+    # Mock HTML for container rows with action dropdowns
+    MOCK_ROWS_HTML = """
+<tr>
+<td>1</td>
+<td data-sort="plex"><a href="/stack/plex" class="link">plex</a></td>
+<td data-sort="server">server</td>
+<td><div class="dropdown dropdown-end">
+<label tabindex="0" class="btn btn-circle btn-ghost btn-xs"><svg class="h-4 w-4"></svg></label>
+<ul tabindex="0" class="dropdown-content menu menu-sm bg-base-200 rounded-box shadow-lg w-36 z-50 p-2">
+<li><a hx-post="/api/stack/plex/restart">Restart</a></li>
+</ul>
+</div></td>
+<td data-sort="nas"><span class="badge">nas</span></td>
+<td data-sort="nginx:latest"><code>nginx:latest</code></td>
+<td data-sort="running"><span class="badge badge-success">running</span></td>
+<td data-sort="3600">1 hour</td>
+<td data-sort="5"><progress class="progress" value="5" max="100"></progress><span>5%</span></td>
+<td data-sort="104857600"><progress class="progress" value="10" max="100"></progress><span>100MB</span></td>
+<td data-sort="1000">↓1KB ↑1KB</td>
+</tr>
+<tr>
+<td>2</td>
+<td data-sort="redis"><a href="/stack/redis" class="link">redis</a></td>
+<td data-sort="redis">redis</td>
+<td><div class="dropdown dropdown-end">
+<label tabindex="0" class="btn btn-circle btn-ghost btn-xs"><svg class="h-4 w-4"></svg></label>
+<ul tabindex="0" class="dropdown-content menu menu-sm bg-base-200 rounded-box shadow-lg w-36 z-50 p-2">
+<li><a hx-post="/api/stack/redis/restart">Restart</a></li>
+</ul>
+</div></td>
+<td data-sort="nas"><span class="badge">nas</span></td>
+<td data-sort="redis:7"><code>redis:7</code></td>
+<td data-sort="running"><span class="badge badge-success">running</span></td>
+<td data-sort="7200">2 hours</td>
+<td data-sort="1"><progress class="progress" value="1" max="100"></progress><span>1%</span></td>
+<td data-sort="52428800"><progress class="progress" value="5" max="100"></progress><span>50MB</span></td>
+<td data-sort="500">↓500B ↑500B</td>
+</tr>
+"""
+
+    def test_dropdown_pauses_refresh(self, page: Page, server_url: str) -> None:
+        """Opening action dropdown pauses auto-refresh.
+
+        Bug: focusin event triggers pause, but focusout fires shortly after
+        when focus moves within the dropdown, causing refresh to resume
+        while dropdown is still visually open.
+        """
+        # Mock container rows and update checks
+        page.route(
+            "**/api/containers/rows/*",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="text/html",
+                body=self.MOCK_ROWS_HTML,
+            ),
+        )
+        page.route(
+            "**/api/containers/check-updates",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="application/json",
+                body='{"results": []}',
+            ),
+        )
+
+        page.goto(f"{server_url}/live-stats")
+
+        # Wait for container rows to load
+        page.wait_for_function(
+            "document.querySelectorAll('#container-rows tr:not(.loading-row)').length > 0",
+            timeout=TIMEOUT,
+        )
+
+        # Wait for timer to start
+        page.wait_for_function(
+            "document.getElementById('refresh-timer')?.textContent?.includes('↻')",
+            timeout=TIMEOUT,
+        )
+
+        # Click on a dropdown to open it
+        dropdown_label = page.locator(".dropdown label").first
+        dropdown_label.click()
+
+        # Wait a moment for focusin to trigger
+        page.wait_for_timeout(200)
+
+        # Verify pause is engaged
+        timer_text = page.locator("#refresh-timer").inner_text()
+
+        assert timer_text == "❚❚", (
+            f"Refresh should be paused after clicking dropdown. timer='{timer_text}'"
+        )
+        assert "❚❚" in timer_text, f"Timer should show pause icon, got '{timer_text}'"
+
+    def test_refresh_stays_paused_while_dropdown_open(self, page: Page, server_url: str) -> None:
+        """Refresh remains paused for duration dropdown is open (>5s refresh interval).
+
+        This is the critical test for the pause bug: refresh should stay paused
+        for longer than the 3-second refresh interval while dropdown is open.
+        """
+        # Mock container rows and update checks
+        page.route(
+            "**/api/containers/rows/*",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="text/html",
+                body=self.MOCK_ROWS_HTML,
+            ),
+        )
+        page.route(
+            "**/api/containers/check-updates",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="application/json",
+                body='{"results": []}',
+            ),
+        )
+
+        page.goto(f"{server_url}/live-stats")
+
+        # Wait for container rows to load
+        page.wait_for_function(
+            "document.querySelectorAll('#container-rows tr:not(.loading-row)').length > 0",
+            timeout=TIMEOUT,
+        )
+
+        # Wait for timer to start
+        page.wait_for_function(
+            "document.getElementById('refresh-timer')?.textContent?.includes('↻')",
+            timeout=TIMEOUT,
+        )
+
+        # Record a marker in the first row to detect if refresh happened
+        page.evaluate("""
+            const firstRow = document.querySelector('#container-rows tr');
+            if (firstRow) firstRow.dataset.testMarker = 'original';
+        """)
+
+        # Click dropdown to pause
+        dropdown_label = page.locator(".dropdown label").first
+        dropdown_label.click()
+        page.wait_for_timeout(200)
+
+        # Confirm paused
+        assert page.locator("#refresh-timer").inner_text() == "❚❚"
+
+        # Wait longer than the 5-second refresh interval
+        page.wait_for_timeout(6000)
+
+        # Check if still paused
+        timer_text = page.locator("#refresh-timer").inner_text()
+
+        # Check if the row was replaced (marker would be gone)
+        marker = page.evaluate("""
+            document.querySelector('#container-rows tr')?.dataset?.testMarker
+        """)
+
+        assert timer_text == "❚❚", f"Refresh should still be paused after 6s. timer='{timer_text}'"
+        assert marker == "original", (
+            "Table was refreshed while dropdown was open - pause mechanism failed"
+        )
+
+    def test_refresh_resumes_after_dropdown_closes(self, page: Page, server_url: str) -> None:
+        """Refresh resumes after dropdown is closed."""
+        # Mock container rows and update checks
+        page.route(
+            "**/api/containers/rows/*",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="text/html",
+                body=self.MOCK_ROWS_HTML,
+            ),
+        )
+        page.route(
+            "**/api/containers/check-updates",
+            lambda route: route.fulfill(
+                status=200,
+                content_type="application/json",
+                body='{"results": []}',
+            ),
+        )
+
+        page.goto(f"{server_url}/live-stats")
+
+        # Wait for container rows to load
+        page.wait_for_function(
+            "document.querySelectorAll('#container-rows tr:not(.loading-row)').length > 0",
+            timeout=TIMEOUT,
+        )
+
+        # Wait for timer to start
+        page.wait_for_function(
+            "document.getElementById('refresh-timer')?.textContent?.includes('↻')",
+            timeout=TIMEOUT,
+        )
+
+        # Click dropdown to pause
+        dropdown_label = page.locator(".dropdown label").first
+        dropdown_label.click()
+        page.wait_for_timeout(200)
+
+        assert page.locator("#refresh-timer").inner_text() == "❚❚"
+
+        # Close dropdown by pressing Escape or clicking elsewhere
+        page.keyboard.press("Escape")
+        page.wait_for_timeout(300)  # Wait for focusout timeout (150ms) + buffer
+
+        # Verify refresh resumed
+        timer_text = page.locator("#refresh-timer").inner_text()
+
+        assert timer_text != "❚❚", (
+            f"Refresh should resume after closing dropdown. timer='{timer_text}'"
+        )
+        assert "↻" in timer_text, f"Timer should show countdown, got '{timer_text}'"

uv.lock

@@ -242,6 +242,7 @@ dependencies = [
 [package.optional-dependencies]
 web = [
     { name = "fastapi", extra = ["standard"] },
+    { name = "humanize" },
     { name = "jinja2" },
     { name = "websockets" },
 ]
@@ -270,6 +271,7 @@ dev = [
 requires-dist = [
     { name = "asyncssh", specifier = ">=2.14.0" },
     { name = "fastapi", extras = ["standard"], marker = "extra == 'web'", specifier = ">=0.109.0" },
+    { name = "humanize", marker = "extra == 'web'", specifier = ">=4.0.0" },
     { name = "jinja2", marker = "extra == 'web'", specifier = ">=3.1.0" },
     { name = "pydantic", specifier = ">=2.0.0" },
     { name = "pyyaml", specifier = ">=6.0" },
@@ -781,6 +783,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/2a/39/e50c7c3a983047577ee07d2a9e53faf5a69493943ec3f6a384bdc792deb2/httpx-0.28.1-py3-none-any.whl", hash = "sha256:d909fcccc110f8c7faf814ca82a9a4d816bc5a6dbfea25d6591d6985b8ba59ad", size = 73517, upload-time = "2024-12-06T15:37:21.509Z" },
 ]
 
+[[package]]
+name = "humanize"
+version = "4.15.0"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/ba/66/a3921783d54be8a6870ac4ccffcd15c4dc0dd7fcce51c6d63b8c63935276/humanize-4.15.0.tar.gz", hash = "sha256:1dd098483eb1c7ee8e32eb2e99ad1910baefa4b75c3aff3a82f4d78688993b10", size = 83599, upload-time = "2025-12-20T20:16:13.19Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/c5/7b/bca5613a0c3b542420cf92bd5e5fb8ebd5435ce1011a091f66bb7693285e/humanize-4.15.0-py3-none-any.whl", hash = "sha256:b1186eb9f5a9749cd9cb8565aee77919dd7c8d076161cf44d70e59e3301e1769", size = 132203, upload-time = "2025-12-20T20:16:11.67Z" },
+]
+
 [[package]]
 name = "identify"
 version = "2.6.15"