A personal AI assistant that operates entirely within GitHub Issues and Actions, eliminating the need for external servers or infrastructure. It targets developers and users seeking an integrated, version-controlled AI workflow, enabling self-updating software projects and persistent conversational memory managed directly within a Git repository.

How It Works

Gitclaw leverages the pi coding agent to process requests submitted via GitHub Issues. Each issue initiates a new AI session, with subsequent comments resuming the conversation with full context. All interactions, including conversation history and file modifications made by the agent, are committed directly to the repository, effectively using Git as a state-storage mechanism. This approach provides a versioned, auditable, and persistent memory for the AI.

Quick Start & Requirements

1. Fork the repository.
2. Add your Anthropic API key as a GitHub Actions secret named ANTHROPIC_API_KEY.
3. Open a new issue in your forked repository to start the agent.
4. Comment on the issue to continue the conversation.

Prerequisites: An Anthropic API key is required. GitHub Actions must be enabled.

Highlighted Details

• Serverless Operation: Runs exclusively on GitHub Actions, requiring no additional hosting.
• Versioned Memory: Conversation history and agent state are stored as Git commits, enabling long-term memory, search, and summarization of past interactions.
• File Manipulation: The agent can read and write files within the repository, facilitating the creation and iterative updating of software projects directly through issues.
• Customization: Workflow behavior (model, tools, thinking depth, triggers) can be configured via .github/workflows/agent.yml.

Maintenance & Community

The project is built upon pi-mono by Mario Zechner. An acknowledgment is made to ymichael for contributing the core idea. No other community links or active maintenance signals are present in the provided documentation.

Licensing & Compatibility

The repository's license is not explicitly stated in the provided text. Public repositories will have all conversation history visible to anyone, while private repositories are recommended for sensitive data. Compatibility is tied to the GitHub Actions environment and the availability of the Anthropic API.

Limitations & Caveats

Conversations in public repositories are publicly visible. The system's functionality is dependent on GitHub Actions' execution limits and availability. Users must provide their own Anthropic API key, incurring associated costs. The effectiveness of the agent is contingent on the capabilities of the underlying pi coding agent.