A self-hosted AI crew that runs 24/7 on your server.
Three specialized agents working autonomously in Docker, built on OpenClaw.

Architecture

%%{init: {'theme': 'base', 'themeVariables': {
  'primaryColor': '#7C3AED',
  'primaryTextColor': '#E0E0FF',
  'primaryBorderColor': '#4F46E5',
  'lineColor': '#A78BFA',
  'secondaryColor': '#4F46E5',
  'tertiaryColor': '#1E1B4B',
  'clusterBkg': '#1E1B4B',
  'clusterBorder': '#4F46E5',
  'edgeLabelBackground': '#1E1B4B',
  'nodeTextColor': '#E0E0FF'
}}}%%
graph TB
    subgraph CrewDock ["CrewDock (Docker)"]
        GW[OpenClaw Gateway]

        GW --> Forge["Forge<br/><i>Dev Autopilot</i>"]
        GW --> Alfred["Alfred<br/><i>Personal Assistant</i>"]
        GW --> Scouter["Scouter<br/><i>Intel Radar</i>"]
    end

    Forge --> |worktrees + PRs| GitHub
    Alfred --> |read/write| Google["Google Workspace"]
    Scouter --> |monitor| Sources["RSS / Twitter / Web"]

    Discord <--> GW

Loading

What is CrewDock

CrewDock turns a Docker host into a 24/7 AI operations center. It runs OpenClaw as the gateway, adds three specialized agents, and wires everything to Discord so you can monitor and interact from your phone.

The agents run on cron schedules or on demand. Each one has its own workspace, config, and database. You deploy once and they take it from there.

The Agents

Alfred — Personal Assistant

Daily briefings and Google Workspace access (Gmail, Calendar, Tasks) via Discord. On first message, Alfred walks you through setting your briefing schedule. After that, it delivers a morning summary on cron and answers workspace queries on demand.

Forge — Dev Autopilot

Autonomous development agent. Picks up GitHub issues, writes code, and opens PRs. Forge uses OpenClaw's ACP (Agent Client Protocol) to spawn isolated coding sessions that run Claude CLI (acpx) inside the container. Each session gets its own worktree, runs autonomously, and reports results back to Discord.

How it works:

1. A cron job fires on a configurable interval (default every 15m, disabled on first boot)
2. Forge checks which repos are due based on their schedule
3. For each repo, it fetches open issues (oldest first)
4. Filters out issues with active sessions, existing PRs, or exclude labels
5. Spawns an ACP session per issue (up to max concurrent) that invokes Claude CLI
6. Each session: reads the issue, creates a worktree, implements, tests, opens a PR
7. Results are announced on Discord

Config example (config.json):

{
  "timezone": "America/New_York",
  "cron": {
    "enabled": false,
    "interval": "*/15 * * * *",
    "jobId": ""
  },
  "defaults": {
    "branch": "main",
    "agentId": "claude",
    "model": null,
    "schedule": "on-demand",
    "thread": true,
    "maxConcurrentSessions": 4,
    "maxAttempts": 3
  },
  "projects": [
    {
      "repo": "your-username/your-repo",
      "branch": "main"
    }
  ]
}

Projects inherit from defaults — only repo is required.

Schedules:

Schedule	Behavior
on-demand	Manual trigger only
always	Every cron cycle
HH-HH	Hour range (e.g., 22-07 wraps midnight)
HH-HH weekdays	Monday–Friday only
HH-HH weekends	Saturday–Sunday only

Global defaults:

Setting	Default	Description
branch	"main"	Base branch for new worktrees
agentId	"claude"	ACP agent for sessions
model	null	Model override (null = agent default)
schedule	"on-demand"	Default schedule for projects
thread	true	Create a Discord thread per session
maxConcurrentSessions	4	Max active autopilot sessions globally
maxAttempts	3	Max retry attempts per issue
cron.interval	"*/15 * * * *"	Cron schedule when enabled

Project options:

Field	Required	Description
repo	Yes	GitHub repo (owner/name)
branch	No	Override default branch
agentId	No	Override default agent
model	No	Override default model
schedule	No	Override default schedule
thread	No	Override default thread setting
enabled	No	Toggle on/off (default: true)
excludeLabels	No	Issue labels to skip
testCommand	No	Custom test command
setupInstructions	No	Run before each session

Scouter — Intel Radar

Monitors AI/tech sources (RSS, Twitter/X, web pages) and drafts engagement posts in your voice for Twitter/X. Never publishes automatically — all drafts go through you on Discord.

Sources config (config.json):

{
  "timezone": "America/New_York",
  "sources": {
    "twitter": {
      "schedule": "twice-daily",
      "list_id": "YOUR_X_LIST_ID",
      "max_results": 10
    },
    "rss": [
      { "name": "HackerNews Best", "url": "https://hnrss.org/best", "schedule": "every-4h" }
    ],
    "web": [
      { "name": "GitHub Trending", "url": "https://github.com/trending", "schedule": "daily-at-10" }
    ]
  }
}

Comes with 8 post templates: build logs, library reviews, news commentary, original takes, quote tweets, replies, resource shares, and threads.

Prerequisites

• Docker and Docker Compose
• curl and jq (used by the install wizard)

The install wizard walks you through everything else: Discord bots, GitHub tokens, Claude credentials, and optional integrations. It validates each credential before saving.

Installation

git clone https://github.com/dasirra/crewdock.git
cd crewdock
./install.sh

The wizard will:

1. Install gum (TUI framework) if not present
2. Let you pick which agents to enable
3. Walk you through each integration (Discord, GitHub, Claude, Google Workspace, X/Twitter)
4. Validate credentials against their APIs in real time
5. Generate your .env and create runtime directories
6. Offer to start the container (make up) and authenticate an LLM provider (make auth)

Power user alternative: skip the wizard entirely.

cp .env.example .env
vim .env        # fill in your values
make up
make auth       # authenticate an LLM provider

Reconfiguring

Run ./install.sh again at any time to add agents, change tokens, or update integrations. The wizard detects your existing .env and lets you modify it.

Configuration

Environment Variables

The install wizard generates .env for you. For manual setup, copy .env.example to .env and fill in your values. See .env.example for the full list of available variables and their descriptions.

Commands

Command	Description
make up	Build and start all services
make down	Stop all services
make restart	Restart all services
make restart-gateway	Restart only the gateway
make logs	Tail gateway logs
make logs-all	Tail all service logs
make status	Show running containers
make auth	Authenticate an LLM provider (interactive selector)
make auth-anthropic	Authenticate Anthropic OAuth
make auth-codex	Authenticate OpenAI Codex OAuth
make version	Show pinned, running, and latest versions
make shell	Open bash shell in the gateway container
make dashboard	Auto-approve pending devices, print dashboard URL
make config-preview	Preview generated openclaw.json without Docker
make clean	Remove dangling Docker images
make help	Show all available commands

Project Structure

crewdock/
├── install.sh                     # TUI installation wizard
├── installer/                     # Wizard modules
│   ├── manifest.json              # Agent and integration definitions
│   ├── lib.sh                     # Shared helpers (output, env, gum wrappers)
│   ├── gum.sh                     # Gum detection and auto-install
│   ├── discord.sh                 # Discord bot setup + validation
│   ├── github.sh                  # GitHub PAT setup + validation
│   ├── claude.sh                  # Claude OAuth token setup
│   ├── gws.sh                     # Google Workspace credentials setup
│   └── xurl.sh                    # X/Twitter API setup + validation
├── agents/                        # Agent templates (tracked in git)
│   ├── forge/                     # Dev autopilot agent
│   ├── alfred/                    # Personal assistant agent
│   ├── scouter/                   # Intel radar agent
│   ├── USER.md                    # User profile shared across agents (gitignored)
│   └── USER.example.md
├── claude/                        # Claude CLI commands
├── init.d/                        # Boot scripts (run on container start)
├── home/                          # Persistent /home/node volume (gitignored)
│   ├── .openclaw/                 # Gateway config + agent workspaces
│   ├── .claude/                   # Claude CLI config
│   └── .config/                   # gh, gws, xurl credentials
├── docker-compose.yaml            # Core service definition
├── docker-compose.override.yaml   # Personal additions (gitignored)
├── Dockerfile                     # Base image + core tools
├── Dockerfile.local               # Personal tool additions (gitignored)
├── docker-entrypoint.sh           # Container entrypoint
├── Makefile                       # Build and management commands
└── .openclaw-version              # Pinned OpenClaw base image version

Extending

Custom Tools

Copy Dockerfile.local.example to Dockerfile.local and add your own tools:

FROM openclaw-openclaw-gateway:latest

USER root
RUN apt-get update && apt-get install -y your-tools
USER node

Additional Services

Copy docker-compose.override.example.yaml to docker-compose.override.yaml and add services. Docker Compose merges it automatically. Both files are gitignored so personal additions won't conflict with upstream updates.

Roadmap

• Tailscale sidecar — secure remote access to the container without exposing ports
• Syncthing + Obsidian vault sync — give agents access to your second brain
• New agents — expand the crew with additional specialized agents

License

MIT