Getting Started

Prerequisites

  • Node.js ≥ 22.5 — the daemon and dashboard server run on it.
  • The claude CLI, logged into your subscription. Run claude once in a normal terminal and /login if needed — the orchestrator drives this real process, it doesn’t replace it.
  • A local LLM to act as the “brain”: LM Studio serving on http://localhost:1234/v1, or Ollama on http://localhost:11434/v1, with a capable instruct model (e.g. a Qwen 30B+).

Install and run

  1. Start LM Studio or Ollama with your chosen model loaded.

  2. Install dependencies (first run only):

    npm run setup

    This installs both the server and the dashboard UI dependencies.

  3. Create your config:

    cp config.example.json config.json

    Set provider.model to a model name your local server reports, and define your session(s): cwd, goal, doneCriteria. See the Configuration reference for every field.

  4. Start it:

    npm start

    This builds the latest dashboard UI, starts the server, and opens http://localhost:4317 in your browser. Click Start all, or start sessions individually.

Windows desktop shortcut: double-click launch.cmd (or run it once to drop an “AGI Dashboard” shortcut on your desktop) to do all of the above with one click — it installs dependencies on first run too. Close the window (or Ctrl+C) to stop.

Prefer a headless console runner instead of the dashboard? npm run daemon runs all sessions and logs the event stream straight to the console.

Everyday scripts

Command Does
npm start / npm run launch Build the UI, serve the dashboard, open the browser.
npm run dashboard Serve the dashboard without rebuilding the UI / opening a browser.
npm run dev Dashboard with tsx watch (auto-restart on server-code changes).
npm run daemon Headless: run every session, log the event stream to console.
npm run build Build the dashboard UI only (web/dist).

Checks (offline — what CI runs)

Command Does
npm run typecheck Type-check the server.
npm run web:check Type-check the UI.
npm test Deterministic unit tests. No claude CLI or LLM needed.

Live validation (needs a claude login + a local model)

Command Does
npm run pty-smoke Prove PTY spawn/read/inject against the real claude.
npm run smoke End-to-end smokes (spin-loop, agent turn, usage).
npm run brain-validate Validate the brain’s decisions against the live model.

Next steps