# gnt.ai

> Git-native rules layer for AI agents. Every rule is a file in your repo, reviewed and merged like code. Agents call check_action before anything risky and get back allowed, blocked, or needs_human, plus the exact rule that made the call.

## Product

- [Homepage](https://gntai.dev/): What gnt.ai is, how it works, and how it stays out of your data.
- [Book a demo](https://gntai.dev/book-demo): Hosted access is offered through guided private pilots; commercial terms are shared directly with participating teams.
- [Changelog](https://gntai.dev/changelog): Release notes will land here shortly.

## Docs

- [Quickstart](https://docs.gntai.dev/docs/quickstart): Get a key, connect an agent, and see how a rule gets approved.
- [Tools & Enforcement](https://docs.gntai.dev/docs/tools): The 5 MCP tools gnt-brain exposes.
- [Connect an Agent](https://docs.gntai.dev/docs/connect): claude.ai, Claude Code, a raw MCP client, OpenAI, Hermes, or OpenClaw.
- [Sources](https://docs.gntai.dev/docs/sources): Pull decision-prose in from the tools your team already uses.
- [API Reference](https://docs.gntai.dev/api): The full HTTP API gnt-brain runs on.

## Source

- [GitHub](https://github.com/gnt-ai/gnt): Apache-2.0 source and self-hosting instructions.