GitAgent

An open-source AI agent framework runs locally, stores behavior and memory in Git, and connects to model providers including Anthropic and OpenAI.

GitAgent is an AI agent framework for developers who want to version an agent's behavior and saved memory alongside its tools. Agent definitions live in a Git repository, where teams can review changes, fork agents and maintain different behaviors on separate branches.

Versioned agents: Git tracks identity, rules, configuration, tools, skills and memory, with diffs and history for each change.
Coding tasks: The agent can read and write files, execute shell commands and save memory for later sessions.
Repository sessions: GitHub repository mode can clone a project, commit and push changes to a session branch, and resume that session later.
Extensions and controls: Custom tools, reusable skills and plugins extend the agent, while lifecycle hooks support action gating and activity logging.
MCP tools: Local or remote MCP servers provide access to tools such as GitHub, Postgres and filesystem services.

GitAgent runs locally and is open source under the MIT license. Developers can use its CLI or embed it in an application through an SDK that streams responses and tool events. A separate execution mode supports a sandbox VM, and voice support is available through an optional package.

Model connections use pi-ai, with providers including Anthropic, OpenAI, Google, xAI, Groq and Mistral. Agent definitions and saved memory live in the repository; requests to hosted model providers require internet access and send model input to the selected service.