An open-source, local-first personal AI agent, inspired by the idea of Meta's Muse (a personal agent that takes work off your plate) and rebuilt for everyone, everywhere, with no region locks and no server holding your life.
Live app: https://aeiouvcode.github.io/open-muse/
- Chat-first. Talking to Muse works like messaging a person. Tell it what needs to get done.
- Goals to plans. State a goal and Muse breaks it into a concrete step plan on the Goals board, then advances it step by step with you.
- Approvals before sensitive actions. A Sentinel policy layer pauses anything that sends, shares, books or buys until you explicitly approve it.
- Memory you control. Muse learns durable facts from conversation, shows you every one, and forgets anything on request - one item or everything.
- Permissions model. Per-app scopes (off / read / read+write) gate what Muse is allowed to plan and draft. Revoke any time.
- Audit trail. Everything Muse has done and plans to do, recorded and inspectable.
- Personal VM. All state lives in your browser. Set a passphrase and the store is encrypted with AES-GCM using a key derived from your passphrase - a key only you hold.
- It is BYO-key: chat runs on your own OpenRouter key, stored in this browser only.
- On static hosting it cannot run after you close the tab; it picks up where it left off when you return and nudges you while it is open.
- Connectors model the permission system locally. They gate planning and drafting; they do not touch your real accounts. External actions are drafts and preparations - the final send is always yours.
It is a single static index.html. Serve it anywhere or open the live Pages link.
No build step, no dependencies, no backend.
MIT - see LICENSE. Built from scratch; no code from Meta or any existing project.