Nanobot AI代理Docker镜像

nanobot-docker

Unofficial community Docker image for nanobot — the lightweight, open-source AI agent for your tools, chats, and workflows.

What This Is

This image bundles nanobot with a set of commonly-needed tools so you can spin up a full-featured AI agent environment with a single docker pull. It is not affiliated with the nanobot project — all credit for nanobot itself belongs to the HKUDS team.

Included Tools

Quick Start

Pull

docker pull ghcr.io/orrinwitt/nanobot-docker:latest

Run

docker run -d \
  --name nanobot \
  -v /path/to/nanobot-data:/root/.nanobot \
  -p 18790:18790 \
  ghcr.io/orrinwitt/nanobot-docker:latest

Only /root/.nanobot needs to be mounted. All tools are included in the image.

Replace /path/to/nanobot-data with a persistent directory on your host. This is where config.json, your workspace, secrets, vault, and any custom scripts (including the optional startup hook) will live.

Docker Compose

An example compose file is included in the repo. The quickest way to deploy:

curl -O https://raw.githubusercontent.com/orrinwitt/nanobot-docker/main/docker-compose.example.yml
docker compose -f docker-compose.example.yml up -d

See docker-compose.example.yml for the full reference.

Pinned Versions

# Pull a specific nanobot version
docker pull ghcr.io/orrinwitt/nanobot-docker:v0.3.5

# Pull a verified stable build
docker pull ghcr.io/orrinwitt/nanobot-docker:v0.3.5-stable

See Releases for the full version history and changelogs.

Version Tags

Volume

Everything persists under /root/.nanobot:

See the nanobot documentation for how to configure providers, models, and tools.

Custom Startup Hook

The entrypoint (baked into the image) checks for a user-provided startup.sh on the mounted volume and runs it before starting nanobot. This lets you add custom startup logic — starting databases, syncing files, launching sidecar services — without modifying the image.

Path inside container: /root/.nanobot/workspace/scripts/startup.sh
Path on your host: <your-volume>/workspace/scripts/startup.sh (where <your-volume> is the directory you mounted at /root/.nanobot)

# Create your startup script on the volume (using the Quick Start mount path)
cat > /path/to/nanobot-data/workspace/scripts/startup.sh << 'EOF'
#!/bin/sh
# Start a database, sync files, launch services, etc.
pg_ctlcluster 17 main start
nextcloudcmd -u user -p pass /root/.nanobot/workspace/vault https://cloud.example.com
EOF

chmod +x /path/to/nanobot-data/workspace/scripts/startup.sh

The script runs with output redirected to /tmp/startup-hook.log inside the container. Failures are non-fatal — nanobot will still start even if the script errors. If the file doesn't exist, the hook is silently skipped.

Tool Configuration

MCP Servers

MCP servers run via npx (cached automatically, no global install). Add them to your config.json:

{
  "mcpServers": {
    "obsidian": {
      "command": "npx",
      "args": ["-y", "@mauricio.wolff/mcp-obsidian", "/root/.nanobot/workspace/vault"],
      "transport": "stdio",
      "disabled": false
    },
    "memory": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-memory"],
      "transport": "stdio",
      "disabled": false
    }
  }
}

Fabric (AI Augmentation Patterns)

257+ patterns are pre-downloaded at build time (no boot-time download delay). Custom patterns persist in your volume at workspace/skills/fabric/patterns/ and are copied into the image on boot.

Configure Fabric via ~/.config/fabric/.env (mount from secrets/fabric.env):

OPENAI_API_KEY=your-api-key
OPENAI_API_BASE_URL=https://your-gateway/api/v1
DEFAULT_MODEL=your-model
DEFAULT_VENDOR=OpenAI
FABRIC_DISABLE_RESPONSES_API=true

# List available patterns
fabric --list

# Use a pattern
echo "content" | fabric --pattern summarize
cat file.txt | fabric --pattern extract_wisdom

See: https://github.com/danielmiessler/fabric

Google Workspace (gws)

The gws CLI provides access to Google services. Authenticate by placing your OAuth credentials at workspace/secrets/gws-auth-user.json — the entrypoint copies them to the right location on boot.

gws auth login
gws gmail list
gws calendar list
gws drive files list

See: https://github.com/googleworkspace/cli

GitHub CLI (gh)

gh auth login
gh issue list
gh workflow run build.yml
gh run list --limit 5

See: https://cli.github.com/

PinchTab (Browser Automation)

PinchTab auto-starts on container boot. Token auth is configured via /root/.pinchtab/config.json in your volume.

# Quick test
curl -X POST http://localhost:9867/navigate \
  -H "Authorization: Bearer <token>" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com"}'

See: https://github.com/pinchtab/pinchtab

Nextcloud Sync

The vault can be synced to a Nextcloud instance using nextcloudcmd. Place your Nextcloud credentials in secrets/credentials.json and add a sync script to workspace/scripts/.

See: https://github.com/nextcloud/desktop

Building from Source

# Build for a specific nanobot version
docker build --build-arg NANOBOT_VERSION=v0.3.5 -t nanobot-docker .

# Build for latest nanobot release
docker build -t nanobot-docker .

Auto-Updates

This repo includes a daily workflow that checks for new releases of:

• nanobot (triggers a new versioned release)
• Fabric (rebuilds latest)
• PinchTab (rebuilds latest)

When a new nanobot version is detected, the workflow:

1. Updates NANOBOT_VERSION in the Dockerfile
2. Commits and pushes to main (triggers a new image build)
3. Creates a git tag and GitHub Release with a link to the upstream release notes

Manual trigger: Actions → "Auto-Update from Upstream" → Run workflow.

License

This Docker image configuration is licensed under the MIT License. See LICENSE.

nanobot is licensed under the MIT License by HKUDS. See their repository for details.

Acknowledgments

• nanobot by the HKUDS team — the core agent framework
• Fabric by Daniel Miessler — AI augmentation patterns
• PinchTab — headless browser automation
• gws — Google Workspace CLI