OpenMuse: Open-Source Personal AI Agent with Browser and Terminal Capabilities

Overview

OpenMuse is a self-hostable, open-source personal AI agent application built with CopilotKit. It is currently in Alpha and designed for self-hosting and building. The agent combines a persistent Chromium browser and an optional Linux workspace, allowing it to browse public pages, run commands in a container, and manage files. The interface supports iOS, Android, and web.

Key Features

• Agent Computer: The optional Linux terminal runs in a nonroot Docker container with a persistent /workspace volume. It includes a terminal for command execution (with 30-second limits). The browser worker is a separate service (Playwright/Chromium) for web interaction. Note: Graphical desktops are listed as future work.
• Web Search: Web search is enabled by default and uses Parallel's free, keyless Search MCP. It sends model-generated queries to Parallel and returns results with source URLs.
• Integrations: Supports Gmail and Calendar via Google OAuth adapters. It also includes features for document processing (PDFs), finance tracking (CSV import), and goal tracking.
• Architecture: The system uses a Hono API, a durable task worker, and CopilotKit Intelligence for conversation persistence. It supports deployment on Render with separate services for the API, web app, and browser worker.

Requirements and Deployment

OpenMuse requires Node 24 LTS and pnpm 11.19.0. A CopilotKit Intelligence project key is required for chat persistence. The project is licensed under MIT, though CopilotKit Intelligence is a separate service. Deployment on Render involves setting up three services: openmuse-api, openmuse-web, and openmuse-browser.