Claw Copilot

Claw Copilot turns raw agent execution into something you can actually follow. Instead of digging through scattered logs, you get a focused dashboard for understanding what ran, when it ran, which subagent it triggered, what tools fired, and what artifacts came out.

✨ Why It Stands Out

• 🧭 Run-aware navigation: inspect OpenClaw at the right level - session -> run -> event
• 🔗 Subagent-aware routing: jump from a parent agent to the exact child session and run it triggered
• 🧹 Cleaner A2A visibility: reduces duplicated sessions_send noise when richer subagent context exists
• 📡 Live execution view: stream updates over SSE while runs are still happening
• 💾 Local-first history: persist sessions and runs in SQLite so refreshes and restarts do not wipe context
• 🌐 Tailscale-first access: open Claw Copilot safely from your phone or laptop without exposing localhost to the public internet

🚀 At A Glance

flowchart LR
    A[OpenClaw session] --> B[Run rail]
    B --> C[Timeline]
    B --> D[Artifacts]
    C --> E[Tool activity]
    C --> F[Model replies]
    C --> G[Subagent jump]
    G --> H[Child session / child run]

🌟 Core Highlights

🧭 Run-aware navigation

Claw Copilot is built around runs, not just chat transcripts.

• Session list with run counts and live status
• Dedicated run rail for each session
• Direct path routing:
  • /claw-copilot/session/:sessionId
  • /claw-copilot/session/:sessionId/run/:runId

🤖 Subagent visibility that actually helps

OpenClaw agent-to-agent flows can get noisy fast. Claw Copilot makes them readable.

• Detects subagent launches from sessions_send
• Preserves parent -> child linkage
• Supports precise jump targets for child runs
• Hides redundant sessions_send timeline noise when a richer subagent event already exists

📜 Timeline + artifacts in one workflow

• Model prompts and replies
• Tool activity and results
• Artifact capture and final reply visibility
• Live updates while runs are still executing

🛡️ Safe local-first storage

• Stores data in local SQLite
• Keeps session/run history across dashboard reloads
• Marks clearly abandoned running runs as interrupted

⚡ Quick Start

Why Tailscale?

Claw Copilot is meant to be useful across devices, not just on the machine running OpenClaw.

• Tailscale gives you a private tailnet URL instead of asking you to expose 127.0.0.1 to the internet
• It works well for phone + laptop access on the same tailnet
• The plugin already knows how to guide setup and expose only the Claw Copilot path

1. Install into OpenClaw from npm

openclaw plugins install claw-copilot
openclaw gateway restart

2. Enable remote access through Tailscale

openclaw claw-copilot remote enable

This command will:

• help install Tailscale if needed
• start Tailscale login if the machine is not connected yet
• expose Claw Copilot to your tailnet
• print the tailnet URL you can open from your phone or laptop
• print QR codes for login and mobile access when available

3. Open Claw Copilot from your phone or computer

1. Install Tailscale on the device you want to use
2. Sign into the same tailnet
3. Open the Claw Copilot URL printed by openclaw claw-copilot remote enable

Once OpenClaw starts running tasks, Claw Copilot begins capturing sessions and runs automatically.

👀 What You Get

📚 Sessions view

• Live session status
• Session labels and metadata
• Pagination for large histories

🏃 Runs view

• Per-session run list
• Run input preview
• Task / tool / artifact counts

🔍 Detail view

• Timeline of execution events
• Artifact panel
• Control actions for stop / pause / redirect

🔗 Routing

Claw Copilot supports direct linking into dashboard state.

• Session route: /claw-copilot/session/:sessionId
• Run route: /claw-copilot/session/:sessionId/run/:runId

This makes it easier to:

• refresh the page without losing context
• share a specific execution view
• jump from a parent agent to the exact child run it triggered

📱 Remote Access

Claw Copilot is designed to be accessed over Tailscale.

Useful commands:

openclaw claw-copilot remote status
openclaw claw-copilot remote enable
openclaw claw-copilot remote disable

The recommended flow is:

1. Run openclaw claw-copilot remote enable
2. Scan the printed QR code or open the login URL if Tailscale needs sign-in
3. Install Tailscale on your phone and join the same tailnet
4. Open the printed Claw Copilot URL

🧪 Development

Run tests

npm test

Type-check + dashboard checks

npm run check

Rebuild after changes

npm run build

If you are developing against a local OpenClaw install, rebuild and reinstall or reload the plugin after changes.

Local development install

openclaw plugins install /path/to/claw-copilot --link
openclaw gateway restart
openclaw claw-copilot remote enable

After that, use the printed tailnet URL from any device on the same tailnet instead of relying on localhost addresses.

📋 Requirements

• OpenClaw >= 2026.3.12
• A Node.js environment capable of building the dashboard and plugin

⚙️ Configuration

openclaw.plugin.json exposes a few plugin settings:

• basePath: dashboard mount path, default /claw-copilot
• dashboardTitle: dashboard title shown in the UI
• imVerbosity: plugin verbosity level

📦 Releases

This repository is set up so that publishing a GitHub Release can publish the package to npm automatically.

Recommended maintainer flow:

1. bump package.json version
2. push the version commit
3. create a GitHub Release with a matching tag like v0.1.0
4. GitHub Actions runs tests, builds the package, and publishes to npm

The release tag must match the version in package.json.

🦞 Current Focus

Claw Copilot is already useful for real OpenClaw debugging and observability workflows, especially when you need to inspect multi-run, multi-agent behavior.

The current focus is simple: make agent execution easier to understand, easier to navigate, and easier to act on.