CopilotKit/openmuse

What it solves

OpenMuse provides a framework for a personal AI agent that can perform complex, multi-step tasks across a browser, a terminal, and personal files. It solves the problem of AI agents being limited to chat interfaces by giving them a "computer" (a persistent browser and optional Linux workspace) to interact with the web and local files, while allowing the user to monitor, review, and take over the process at any time.

How it works

OpenMuse operates using a distributed architecture consisting of a client (React Native/Web), an API server (Hono + CopilotKit runtime), and specialized workers:

• Browser Worker: A Playwright-based Chromium service with persistent profiles that allows the agent to browse public pages and handle interactive web tasks.
• Computer/Terminal: An optional isolated Linux container where the agent can run commands and manage files in a persistent workspace.
• Task Worker: A durable engine that manages task plans, progress, and approvals, using SQL leases to recover interrupted work.
• Integrations: Adapters for Google OAuth (Gmail and Calendar) to allow the agent to read and write personal data with user review.

Who it’s for

It is designed for developers who want to build or self-host a personal AI agent capable of autonomous web browsing, file manipulation, and personal productivity automation across iOS, Android, and web platforms.

Highlights

• Agent Computer: Combines a persistent Chromium browser and a Linux workspace for real-world interaction.
• Human-in-the-loop: Users can review actions, approve steps, and use "Take control" to jump directly into the agent's active browser session.
• Cross-Platform: Built with CopilotKit React Native for native mobile and web support.
• Personal Context: Integrates with Gmail and Google Calendar for managing emails and events.
• Durable Tasks: Supports long-running tasks with the ability to pause, resume, cancel, or retry.