Project Cerebro - Control Surface

A highly modular, agent-friendly execution control surface for running multiple OpenClaw “brains” (agents) with:

• an Execution Stream (source of truth)
• Recurring Tasks that spawn instances into the stream
• per-brain Auto Mode + manual Run
• channel-agnostic notifications via OpenClaw messaging

!Dashboard screenshot

What this is

Cerebro is a dashboard + runtime that helps you supervise a set of specialized agents (“brains”).

Core model

• Execution Stream = source of truth for everything that will run / has run.
• One-time tasks execute when you click Run or when the brain is in Auto Mode.
• Recurring tasks spawn exactly one instance into the Execution Stream when due and execute regardless of Auto Mode.

Brains

• A brain is a named scope + tool/context configuration bound to an OpenClaw agent id.
• Nexus is the default brain: general-purpose orchestration and “everything else.”

Install / Setup (high level)

See SETUP.md for full instructions.

Required secrets / env vars

• OPENCLAW_GATEWAY_URL (e.g. ws://127.0.0.1:18789)
• OPENCLAW_TOKEN (OpenClaw Gateway token)

Quick outline:

1) Clone repo + install deps

2) Create .env from .env.example and set the required OpenClaw vars above

3) Copy templates:

• config/brains.template.json → config/brains.json
• config/brain_targets.template.json → config/brain_targets.json

4) (Optional) seed examples:

• npm run seed

Modularity (agent-friendly)

This repo is designed for agents to customize quickly:

• Add/remove brains by editing config/brains.json and config/brain_targets.json.
• Optional Job module: opt in/out by adding/removing the Job brain + pages (see config/brains.jobs.addon.template.json).
• Recurring tasks are templates under config/seeds/.

Runtime behaviors (security transparency)

This package runs a local HTTP service (Fastify + React UI) and will:

• Expose a dashboard over HTTP on PORT (default 3000)
• Call the OpenClaw Gateway/CLI using the secrets you provide
• Send outbound messages using OpenClaw messaging to the per-brain {channel,target} destinations
• Accept file uploads via POST /api/upload and store them locally under ./data//intake/

Notes

• No secrets or personal IDs are committed; local config/DB are gitignored.
• Notifications are routed through OpenClaw’s channel system (Discord/Telegram/Signal/etc.).
• Optional modules (Job tracker/profile) are opt-in; if you don’t include the Job brain, the UI hides job pages.