Run coding agents in the background to generate pull requests while you work
Issue instructions from the web, Slack, GitHub, or Linear, and the agent writes code in a sandbox and creates a pull request. This open-source background coding agent system is inspired by Ramp's Inspect. Multiple people can join the same session to view results in real time and collaborate.
Scope of agent tasks
The agent has access to the entire development environment, including Node.js, Python, git, browser automation, and VS Code.
- Support for various AI models You can choose from Anthropic Claude, OpenAI Codex, xAI Grok, or OpenCode Zen. You can also use models without API keys by leveraging existing subscriptions.
- Real-time multiplayer sessions Multiple users can join the same session to exchange prompts, and git commits record who wrote which code.
- Automation and scheduling You can set up automations for recurring cron jobs or reactions to external events, which automatically pause upon failure.
Security structure to verify before deployment
This system is designed for single-tenant use only, so it is safe only in environments where all users within the organization are trusted. Access to team-owned sessions requires team membership, and deletion permissions are granted only to the session owner, team lead, or workspace administrator. Credentials are issued when needed but are cached on disk, so be aware that they may be included in snapshots. If you want a multi-tenant deployment, you need per-tenant GitHub app installations and data model isolation.
Suitable for internal team automation development
This tool is suitable if a trusted internal team wants to handle repetitive coding tasks or event-based automation. Before deployment, you must run it behind SSO/VPN, minimize the repository access scope of the GitHub app, and configure restrictions on allowed GitHub users or organizations. It is released under the MIT license, allowing you to modify and distribute it freely.
