Luciano Castillo
8db30f1f12
* refactor(clawdbot): update spawn inference logic - Replaced parent session activity tracking with a pending spawns array to record Task tool calls. - Adjusted the spawn inference window to 10 seconds, ensuring accurate tracking of recent Task calls. - Enhanced the inferSpawnedBy function to utilize the new pending spawns structure for improved session mapping. - Removed obsolete parent session activity tracking code for cleaner implementation. * fix(clawdbot): normalize tool name handling in action tracking - Updated the action tracking logic to convert tool names to lowercase for consistent comparison, ensuring accurate detection of 'Task' tool calls in both addAction and hydrateFromServer functions. - Enhanced event parsing to process tool events more effectively, improving overall event handling and session state management. * refactor(clawdbot): enhance spawn inference with parent session activity tracking - Introduced a new Map to track recent parent session activity, providing a fallback for spawn inference when Task tool calls are not available. - Updated the recordPendingSpawn and inferSpawnedBy functions to utilize this new tracking mechanism, improving the accuracy of session mapping. - Enhanced the addAction and hydrateFromServer functions to incorporate parent session activity tracking, ensuring comprehensive spawn history management. - Cleared parent session activity during collection resets to maintain data integrity. * refactor(clawdbot): improve parent session action tracking for spawn inference - Replaced the previous parent session activity tracking with a new structure to maintain a history of recent actions, allowing for more precise spawn inference. - Updated the trackParentAction function to log actions with timestamps and manage a maximum history length. - Modified the inferSpawnedBy function to utilize the new action history for better accuracy in linking subagents to their parent sessions. - Adjusted the addAction and hydrateFromServer functions to reflect the new tracking mechanism, ensuring comprehensive session mapping. - Cleared the action history during collection resets to maintain data integrity. * style(monitor): adjust ActionGraph component dimensions - Updated the ActionGraphInner component to include width and height styles, enhancing the layout and visual consistency of the graph representation.
🦀 Crabwalk
Real-time companion monitor for Moltbot (Clawdbot) agents by @luccasveg.
Watch your AI agents work across WhatsApp, Telegram, Discord, and Slack in a live node graph. See thinking states, tool calls, and response chains as they happen.
Features
- Live activity graph - ReactFlow visualization of agent sessions and action chains
- Multi-platform - Monitor agents across all messaging platforms simultaneously
- Real-time streaming - WebSocket connection to clawdbot gateway
- Action tracing - Expand nodes to inspect tool args and payloads
- Session filtering - Filter by platform, search by recipient
Installation
Docker (recommended)
docker run -d \
-p 3000:3000 \
-e CLAWDBOT_API_TOKEN=your-token \
-e CLAWDBOT_URL=ws://host.docker.internal:18789 \
ghcr.io/luccast/crabwalk:latest
Note: When running Crabwalk in Docker, the Moltbot gateway typically runs on the host. Use
CLAWDBOT_URL=ws://host.docker.internal:18789so the container can connect. If you're running Moltbot withbind: loopbackandtailscale servefor secure tailnet-only access, you'll need to run the crabwalk container with host networking - replacep:3000:3000with--network hostThis allows the container to reach 127.0.0.1:18789 while maintaining the security benefits of loopback-only binding.
Or with docker-compose:
curl -O https://raw.githubusercontent.com/luccast/crabwalk/master/docker-compose.yml
CLAWDBOT_API_TOKEN=your-token CLAWDBOT_URL=ws://host.docker.internal:18789 docker-compose up -d
If gateway is
bind: loopbackonly, you will need to edit thedocker-compose.ymlto addnetwork_mode: host
From source
git clone https://github.com/luccast/crabwalk.git
cd crabwalk
npm install
CLAWDBOT_API_TOKEN=your-token npm run dev
Open http://localhost:3000/monitor
Configuration
Requires clawdbot gateway running on the same machine.
Gateway Token
Find your token in the clawdbot config file:
# Look for gateway.auth.token
cat ~/.clawdbot/clawdbot.json | rg "gateway\.auth\.token"
Or with jq:
jq '.gateway.auth.token' ~/.clawdbot/clawdbot.json
Or copy it directly:
export CLAWDBOT_API_TOKEN=$(python3 -c "import json,os; print(json.load(open(os.path.expanduser('~/.clawdbot/clawdbot.json')))['gateway']['auth']['token'])")
Accessing from a remote host
If you are running this on a remote server and accessing it through a non-local browser, the default allowedHosts behaviour will prevent access to the web UI.
If running in docker, you can pass the environment variable ALLOWED_HOSTS with a comma-separated list of hosts you wish to allow access to the crabwalk UI.
If running from source, you can either pass ALLOWED_HOSTS as an env var at the command line, or use a .env file.
Stack
TanStack Start, ReactFlow, Framer Motion, tRPC, TanStack DB