Mouse-1

Mouse is a lightweight platform that runs your repository in a cloud sandbox and reports changes for review. It focuses on keeping agents active without needing local setups or accounts.

You start by pushing your repo to Mouse. The service boots a Firecracker microVM that boots in 125 ms and runs at about 95 % of bare‑metal speed. Within that isolated environment your code builds, tests, and runs. When a build succeeds the results appear in a mobile‑friendly timeline on your phone, where you can review outputs, approve the next step, or send the agent back to work. If the agent stalls, a notification tells you that a decision is required. You can then tap a plan or directly answer the question from your phone.

Mouse works best for developers who want to interact with an autonomous agent from a mobile device. The app bundles are kept under 50 KB, making it useful for mid‑range phones. You can choose a language model for each task: Claude Opus 5, ChatGPT 5.6, Sol, or Kimi K3. There is an “Auto” option that picks a model automatically. The timeline shows plan proposals from the agent in a “Plan Mode” that lets it think before it builds, giving you a chance to refine or approve the plan. The built‑in editor keeps your repo, terminal, and all context in one place, while the cloud workspace stays dedicated to the agent.

Mouse offers an API endpoint that allows third‑party tools to invoke its core functions. There is no backing server required for small demos such as a static transit map page that uses a JSON fixture. The demo demonstrates pinch‑zoom on an SVG map rendered by the agent, with the logic seed shipped from a fixture. The entire bundle stays under 50 KB.

Key features

• MicroVMs that boot in 125 ms and run at 95 % bare‑metal performance
• Mobile‑timeline interface to review plans, approve changes, and handle blockers
• Ability to pick or auto‑select language models for each task
• Integrated repository, terminal, and context view for simplified development
• SMS‑style notifications when an agent needs a decision or finishes a task
• No local backend or account required for small static demos