Claude Orchestrator

A desktop app that runs the Claude CLI for you — across many projects, around the clock.

You give it your projects and a plan.md for each. It works through the plan one task at a time, running a separate Claude session per task. When Claude hits a usage limit, the app waits and automatically resumes when the limit resets. When Claude needs a decision (a permission or a question), it appears in a dashboard inbox for you to answer — and the session continues instantly.

Why this exists. The claude CLI stops when it hits a usage limit and never comes back on its own; it also treats each task as a one-off and won't move on to the next item in your plan. This app is the "manager" that sits on top of it.

---

Is this going to cost me money?

No extra money. The app drives the same claude command you already use in a terminal, signed in with your subscription (Pro/Max). It never uses the paid per-token API. The only real limit is your subscription's usage window (a rolling ~5-hour limit plus a weekly cap) — the app is designed to respect and wait those out, not to get around them.

---

Quick start (5 minutes)

You need Node.js 20+ and pnpm, plus the claude CLI installed and logged in (claude → sign in once).

pnpm install      # install dependencies (downloads Electron the first time)
pnpm dev          # launch the app with hot-reload

A window opens. On first run it checks that claude is installed and logged in and shows you the result. That's the Phase 0 scaffold — the full dashboard is being built out phase by phase (see docs/plan and the task list).

Other commands:

pnpm typecheck    # check types (run this before committing)
pnpm test         # run unit tests
pnpm build        # compile a production build into ./out
pnpm package      # build a Windows installer into ./dist

---

New here? Read these in order

The docs/ folder is written for someone new to Electron and to this project. Start at the top:

1. docs/01-getting-started.md — set up your machine and run the app.
2. docs/02-architecture.md — how the app is put together (the three-process Electron model, in plain English).
3. docs/03-how-orchestration-works.md — how we actually drive Claude: sessions, questions, and usage limits.
4. docs/04-contributing-guide.md — how to make your first change, where code lives, and our conventions.
5. docs/05-glossary.md — every unfamiliar word, defined.
6. docs/06-licensing.md — the MIT/Apache-only rule and how to keep the app commercially usable.

---

Tech stack

• Electron — turns a web app into a native desktop app.
• React 18 + Fluent UI v9 — the user interface (same UI toolkit as the vipper-iam and ee.manager projects).
• TypeScript (strict) everywhere.
• The claude CLI — driven as an external subprocess over its stream-json protocol (we do not bundle Anthropic's SDK; see docs/06-licensing.md for why).
• electron-vite for fast builds/hot-reload, Vitest for tests, electron-builder for packaging.

All bundled dependencies are permissive (MIT/Apache/BSD-style) so the app can be used commercially. See docs/06-licensing.md.