GitHub radar

OpenMuse: A Personal AI Agent With Its Own Browser

CopilotKit released OpenMuse, a personal agent with its own browser, terminal, and files that carries a task through to the end instead of just suggesting what to do.

01CopilotKit/openmuse 3.2kTypeScript

OpenMuse is a personal agent with its own browser, terminal, and file access, which CopilotKit describes as carrying work through to a result rather than just suggesting one. It runs its own server, task worker, and browser worker, and can be self-hosted or deployed with one click on Render. In the team's own demo, the agent reads a school-trip email, looks up matching aquarium exhibits, fills out a form from a PDF attachment, and prepares a reply for review; on mobile, it pulls interesting stories from Hacker News and summarizes them. There is a separate module for watching web pages for changes such as a price drop, and one for turning an imported transaction CSV into a spending summary. The optional terminal runs inside an isolated Docker Linux container with no host file access and no outside network beyond the browser worker.

Why a vibe-coder should care

I solve development tasks with Claude Code, and this is a different genre - an agent for personal chores like mail, browsing, and files. The idea is interesting, but this is an alpha and a template you build on yourself, not a finished app you install in five minutes. A CopilotKit Intelligence key and a model provider key (OpenAI, Anthropic, or Google) are required in every mode, so there is no free, subscription-free way to run it, and that is a real downside. I would not rush to give it OAuth access to a real work mailbox, but trying it on a test account to see the pattern seems reasonable.

How to install

Copy this and send it to your agent — Claude Code, Codex, any of them:

Deploy OpenMuse from https://github.com/CopilotKit/openmuse — follow the README, ask me for a CopilotKit Intelligence key and a model provider key (OpenAI, Anthropic, or Google), start it, and show me the agent completing a test task in chat.

Runs on a regular laptop with Node.js — but your own CopilotKit Intelligence key and a model provider key (OpenAI, Anthropic, or Google) are required, so it is neither free nor offline.

Open on GitHub