TypeScript and Python SDK giving AI agents 1000+ pre-authenticated app toolkits, per-user sessions, and a sandbox to take actions.
open-gitagent/gitagent
A git-native AI agent framework where an agent's identity, rules, memory, tools and skills live as version-controlled files in a repo.
- Stars
- 720
- Forks
- 138
- Open issues
- 60
- Last commit
- 8 hours ago
What is it?
Gitagent treats an AI agent as a git repository. Its personality (SOUL.md), behavioral rules (RULES.md), memory (memory/), declarative YAML tools, composable skills and lifecycle hooks are all plain files, so an agent can be forked, branched, diffed and inspected with ordinary git commands.
It ships as a CLI (`gitagent`) and an SDK exposing a streaming `query()` function. The CLI can run against a local directory or clone a GitHub repo, working on a session branch that is auto-committed and pushed. Model access goes through pi-ai, covering providers such as Anthropic, OpenAI, Google, xAI, Groq and Mistral, with fallback models configurable in agent.yaml.
The framework also acts as an MCP client, registering tools from configured MCP servers over stdio or HTTP, and includes plugins, sub-agent delegation, compliance settings with audit logs to .gitagent/audit.jsonl, and optional OpenTelemetry tracing.
Key features
- 01Agent identity, rules, memory and tools are stored as git-versioned files.
- 02Streaming SDK via query() with hooks for gating and modifying tool calls.
- 03Runs against a local directory or clones a GitHub repo onto a session branch.
- 04Acts as an MCP client, auto-registering tools from configured servers.
- 05Supports declarative YAML tools, skills and plugins.
- 06Optional OpenTelemetry tracing and audit logging.
Use cases
- Maintaining a personal or team AI assistant whose rules and memory are reviewed through git diffs.
- Forking a base agent and branching personality or rule variants.
- Running an agent against a GitHub repo to make fixes on a session branch.
- Embedding an agent in a Node.js app via the query() SDK with custom tools.
Quick start
- 1
Install the CLI and SDK via npm
npm install -g @open-gitagent/gitagent - 2
Set your OpenAI API key
export OPENAI_API_KEY="sk-..." - 3
Run your first agent on a project directory
gitagent --dir ~/my-project "Explain this project and suggest improvements"
Things to know
- You need an API key from a model provider such as OpenAI or Anthropic to run agents.
- Voice mode and the web UI are a separate package, @open-gitagent/voice, and must be installed on their own.
Topics
FAQ
- What is gitagent?
- A git-native AI agent framework where an agent's identity, rules, memory, tools and skills live as version-controlled files in a repo. Gitagent treats an AI agent as a git repository. Its personality (SOUL.md), behavioral rules (RULES.md), memory (memory/), declarative YAML tools, composable skills and lifecycle hooks are all plain files, so an agent can be forked, branched, diffed and inspected with ordinary git commands.
- Is gitagent free to use? What license does it have?
- gitagent is open source under the MIT license. Check the license terms before using it in commercial products.
- Is gitagent still maintained?
- Actively maintained. The last commit to gitagent was on Oct 22, 2026.
- What can I use gitagent for?
- Maintaining a personal or team AI assistant whose rules and memory are reviewed through git diffs. Forking a base agent and branching personality or rule variants. Running an agent against a GitHub repo to make fixes on a session branch. Embedding an agent in a Node.js app via the query() SDK with custom tools.
- How do I install gitagent?
- Install the CLI and SDK via npm: npm install -g @open-gitagent/gitagent Set your OpenAI API key: export OPENAI_API_KEY="sk-..." Run your first agent on a project directory: gitagent --dir ~/my-project "Explain this project and suggest improvements"
- Where can I get gitagent?
- The source code and installation instructions are on GitHub: https://github.com/open-gitagent/gitagent
Summary generated by AI from the project's README. GitHub stats refresh daily.
Related projects
The official TypeScript SDK for building Model Context Protocol servers and clients that give LLM applications standardized context.
Open-source SDK for building and running AI agents in Python and TypeScript, with tools, MCP, memory, and model portability.
Open-source shared workspace where humans and AI agents work in one pod, each agent keeping its own memory, skills and identity.
An open-source agent harness in TypeScript for building stateful LLM agents, with a built-in coding agent, CLI, Web UI, and SDKs.
A Go implementation of the Model Context Protocol for building MCP servers that expose tools, resources and prompts to LLM applications.