The open-source AI coding agent for local development.
wuu works directly in your local repository. It can read and edit files, run commands, review changes, and carry a task through to completion using a model provider you configure.
Use the desktop app for interactive work. For larger tasks, wuu can make a plan, delegate work to subagents, use task-specific skills, and continue from persistent sessions. wuu exec provides a non-interactive entrypoint for scripts, CI, and other agents.
Warning
wuu is an early preview and changes quickly. Packaged desktop builds currently support Apple silicon Macs.
Download the latest macOS desktop build from GitHub Releases, move wuu.app to /Applications, and open it.
The current preview is unsigned and not notarized. If macOS blocks an official release that you trust, run:
xattr -dr com.apple.quarantine /Applications/wuu.app && open /Applications/wuu.app- Open
wuu.appand choose a local project folder. - Open Settings → Model providers and connect Anthropic or an OpenAI-compatible provider.
- Start a conversation and describe the outcome you want.
For example:
Explain how this repository is organized.
Fix the failing tests and verify the result.
Review the current changes for regressions.
See the user guide for provider setup, attachments, sessions, permissions, and common problems.
- Work in your repository — read and edit files, search code, run commands, and inspect the resulting diff.
- Handle longer tasks — plan multi-step work, track durable goals, and continue across context limits.
- Delegate work — use subagents for focused research, parallel work, or isolated implementation.
- Keep useful context — resume sessions, fork from a checkpoint, and use persistent memory and skills.
- Bring your own model — connect Anthropic or any OpenAI-compatible endpoint, including local gateways.
- Use files and images — attach documents and screenshots when they are part of the task.
- Automate workflows — stream structured output to scripts, CI jobs, review tools, and other agents.
wuu exec exposes the agent runtime as a non-interactive command:
wuu exec --json "review the current diff"
wuu exec --file plan.md "implement this plan"
wuu exec review --uncommittedSee wuu exec for setup, JSONL events, attachments, session controls, and review options.
- You choose the model provider and credentials. Prompts and relevant context are sent to that provider.
- Provider settings and API keys stay under user control rather than coming from the repository.
- Sessions, config, logs, and other local state live under
~/.wuuby default. - File changes and commands run locally within the selected workspace and active permission mode.
Read the security model before using wuu with untrusted repositories or sensitive data.
- User guide
wuu execautomation guide- Security model
- Roadmap
- Public evaluations
- Changelog
- Documentation index
Contributions are welcome. See CONTRIBUTING.md for development and review guidelines, and SECURITY.md for vulnerability reports.
If you run into a problem, open an issue.
wuu draws inspiration from these projects and their work on coding agents, tool loops, orchestration, and developer experience: