OpenMuse icon

OpenMuse

OpenMuse is an open-source personal agent for iOS, Android, and web that combines chat with browser, terminal, file, document, planning, and goal-tracking workflows. It is intended for people who want an agent that can continue work across sessions and review important actions before they are applied.

OpenMuse

An open-source personal agent for ongoing work

OpenMuse is an open-source personal agent for iOS, Android, and the web. It is designed to handle outcomes rather than only answer questions: users can ask for work, follow the agent's plan, review important actions, and return to results after an interrupted or closed session.

The app brings together streamed chat, a persistent browser and optional terminal, files and document handling, durable task activity, goals, ideas, Gmail and Calendar actions, and editable personal context. It is built with CopilotKit and AG-UI, and is presented as an Alpha template that developers can adapt to their own frontend and agent harness.

OpenMuse also demonstrates a learning workflow through CopilotKit Intelligence. The system can learn from agent interactions and propose reusable skills, but proposed skills require approval before they are added to the agent.

Core capabilities

Browser and terminal workspace

The agent can use a persistent Chromium profile and an optional Linux terminal. Users can reopen the same session and take control when needed.

Persistent task activity

Durable plans show progress, approvals, and receipts, with controls to pause, resume, cancel, or retry interrupted work.

Streamed chat and rich results

Chat supports streamed AG-UI interactions and inline cards for email, browser activity, PDFs, plans, and finance results. Drafts survive a stopped interaction.

Document workflows

The app can work with email attachments and PDFs, including viewing, filling fields, reviewing, and sharing a returned copy.

Reviewed Gmail and Calendar actions

Gmail threads, drafts, and calendar changes are available through Google OAuth, while sends and event changes remain subject to review.

Goals and personal context

Goals can include milestones and recurring checks for public-page changes, availability, or prices crossing a chosen threshold; personal context such as name, tone, avatar, and memories is editable.

Practical uses

  • Managing ongoing tasks

    Ask the agent to carry out a multi-step task, inspect its plan and approvals, then pause, resume, retry, or review receipts as the work progresses.

  • Browser-based research and operations

    Use a persistent browser session and optional terminal for web-based work, then reopen the same computer context and take control instead of starting over.

  • Preparing document responses

    Process an email attachment into a filled PDF, review the completed document, and share a copy when the result is ready.

  • Reviewing communications and scheduling

    Review email threads and drafts or request calendar changes through Google OAuth, with human review before sending or applying changes.

  • Tracking goals and monitored conditions

    Set milestones or recurring public-page checks for availability, changes, or price thresholds, while keeping personal preferences and memories editable.

Pros and Cons

Pros

  • Open-source MIT project with a modifiable template and support for iOS, Android, and web.
  • Combines conversation, browser and file work, task planning, document handling, goals, and personal context in one agent interface.
  • Durable activity, approvals, receipts, pause/resume controls, and review gates make ongoing work inspectable.
  • AG-UI support and CopilotKit headless hooks allow the frontend and agent harness to be swapped or adapted.
  • The learning flow proposes skills from interaction evidence and requires approval before ingestion.

Cons

  • The project is labeled Alpha, so production readiness and operational scope are not established by the source.
  • Some capabilities depend on optional services or accounts, including model access, Google OAuth, and CopilotKit Intelligence for the connected experience.
  • The source describes the app and local setup in detail but provides limited evidence about deployment operations, team administration, and broader integration coverage.

FAQ

What is OpenMuse?

OpenMuse is an open-source personal agent for iOS, Android, and web. It combines chat with a browser, optional terminal, files, task plans, documents, goals, and personal context.

Do I need a model key or external accounts to try it?

The sample app can be run locally without a model key, Google account, Docker, or CopilotKit Intelligence subscription. Those services can be added when needed for a connected deployment.

Can OpenMuse send email or change my calendar without approval?

No. Email sends and calendar changes through the Gmail and Calendar surfaces wait for the user's review before they are applied.

Is OpenMuse ready for production?

The project is labeled Alpha. Its local sample can be started with the documented Node 24 and pnpm 11 commands, and the site presents it as a template that can be modified, rather than as a finished production service.

Can developers customize the app and agent harness?

OpenMuse uses CopilotKit headless hooks and speaks AG-UI. Its sample repository includes a mobile app, Hono API, browser worker, and optional Linux computer service, allowing developers to change the interface or agent harness.

Quick Facts

Category
AI agent
Platforms
iOS, Android, and web
License
MIT open source
Project status
Alpha
Protocol and framework
AG-UI and CopilotKit
Local sample
Node 24 and pnpm 11