GitClaw is an open-source framework for building AI agents whose entire identity, configuration, memory, and capabilities live inside a Git repository. Instead of storing agent state in databases or application code, the framework treats a repository itself as the agent’s environment, allowing developers to version, inspect, and collaborate on agents using standard Git workflows. The system defines structured files that represent the agent’s personality, rules, configuration, and operational logic, enabling transparent control over how the agent behaves. For example, identity and personality may be defined in files such as SOUL.md, while behavioral constraints and policies can be placed in rule definitions. Memory is persisted directly in the repository as version-controlled files, which means conversations, experiences, or learned data can be tracked over time using Git history.
Features
- Git-native agent architecture where the entire agent lives in a repository
- Version-controlled identity, rules, memory, tools, and configuration files
- Declarative tool and skill system for extending agent capabilities
- Persistent memory stored in Git with full history tracking
- Lifecycle hooks that allow scripts or programs to run during agent events
- Collaborative development using standard Git workflows such as pull requests