Rig

A dispatch console for the pi coding agent. Dispatch work from your phone, watch it happen in real time, check back later.

Rig wraps pi --mode rpc as a web interface served over Tailscale — your personal operations dashboard for AI-assisted coding.

What it looks like

• The Board — a flat list of all sessions across all projects, newest first. Active sessions pulse. Tap to open.
• Session Log — a work log, not a chatbot. Tool calls are first-class log entries (read, edit, bash, write) with markdown-rendered agent prose/thinking interspersed.
• New Dispatch — modal for sending work: pick a project, type a prompt, choose a model, choose thinking level (when supported), go. Under 5 seconds.

Quick start

Prerequisites: Node.js 20+, pi installed globally (npm i -g @mariozechner/pi-coding-agent).

# Development (two terminals)
cd server && npm install && npm run dev    # Fastify on :3100
cd frontend && npm install && npm run dev  # Vite on :5173

# Production (top-level supervisor)
./start.sh   # builds frontend/server/operator, serves Rig on :3100, operator REST on :3200

Open http://localhost:5173 (dev) or http://localhost:3100 (production).

Architecture

browser ←→ Vite/static ←→ Fastify ←→ pi --mode rpc (child process)
              :5173          :3100        stdin/stdout JSON-RPC

• Frontend — React 19, TypeScript, Tailwind CSS 4, Vite. Mobile-first responsive layout.
• Server — Fastify 5, WebSocket proxy, reads pi's session files and config directly from ~/.pi/agent/.
• No pi imports — the server always spawns pi --mode rpc and talks over stdin/stdout. Whatever version of pi you have installed is what Rig uses.

Key endpoints

Method	Path	Description
GET	/api/sessions	List all sessions (optional ?cwd= filter)
GET	/api/sessions/:id/entries?path=	Read parsed JSONL entries for a session file
POST	/api/dispatch	Start a new session { cwd, message, provider?, model?, thinkingLevel? }
POST	/api/resume	Resume an existing session { sessionFile, cwd }
POST	/api/stop	Kill an active session { bridgeId }
WS	/api/ws/:bridgeId	Live event stream + command/response channel
GET	/api/models	Enabled models + default from pi settings
GET	/api/models/all	All available models from pi runtime
GET	/api/models/capabilities	Thinking levels for { provider, modelId }
GET	/api/projects	Registered + auto-discovered project directories
GET	/api/browse	Directory browser data for project picker

Configuration

Rig stores its config at ~/.pi/agent/rig.json:

{
  "port": 3100,
  "operator": {
    "telegram": {
      "botToken": "123:abc",
      "allowedChatIds": [123456789]
    },
    "defaultModel": {
      "provider": "openai",
      "modelId": "gpt-5"
    }
  },
  "projects": [
    { "path": "/home/you/projects/my-app", "name": "my-app" }
  ]
}

Projects are also auto-discovered from pi's session history — any directory where you've previously run pi will appear automatically.

Model configuration (enabled models, default model, API keys) is read from pi's own ~/.pi/agent/settings.json. The optional operator section is shared with the Telegram/REST operator process. operator.defaultModel controls the operator assistant's own conversation model. It does not automatically select the model for dispatched Rig coding sessions; those still require an explicit model choice.

Deployment

The ops/ directory has service configs for running Rig as a daemon:

• macOS — ops/com.lolwierd.rig.plist and ops/com.lolwierd.rig-operator.plist (launchd)
• Linux — ops/rig-server.service and ops/rig-operator.service (systemd user services)

Use ./deploy.sh to build frontend/server/operator, sync artifacts into ~/rig-deploy/, install unit files, and restart services.

ops/rig.service is a legacy combined unit kept for reference.

Intended to run on a home server or dev machine accessible over Tailscale, so you can dispatch work from anywhere.

Design

See DESIGN.md for the full design rationale — typography choices (Bricolage Grotesque + IBM Plex Mono), the warm charcoal + amber color palette, and why this is a work log instead of a chatbot.

Stack

Layer	Tech
Frontend	React 19, TypeScript, Tailwind CSS 4, Vite 7
Icons	lucide-react
Server	Fastify 5, @fastify/websocket, @fastify/static
Agent	pi (@mariozechner/pi-coding-agent) via RPC
Runtime	Node.js 20+

License

Personal project. Not published.