gnt.ai — A git native rules layer for AI agents to…

Pitch

Description

Comments

About this project

gnt.ai transforms how AI agents manage rules with its git-native rulebook. By integrating rules as code files reviewed like any other code, teams gain clarity and accountability. Each agent check is recorded, ensuring every action taken has a traceable history, leading to safer decision-making in your operations.

gnt: A Git-Native Rules Layer for AI Agents

The gnt project introduces a groundbreaking approach to managing rules for AI agents using Git. With gnt, every rule functions as a standalone file, integrating seamlessly into the development workflow with the same review and approval processes that developers rely on. This innovative solution ensures enhanced traceability and accountability for actions taken by AI agents.

Key Features

Git-Native Rulebooks: Rules are stored as files in a Git repository and are managed through standard pull requests (PRs), eliminating the need for complicated dashboards or manual approval processes.
Action Verification: Before executing critical actions—such as processing refunds or deleting data—agents utilize the check_action function. This step verifies compliance with the established rules and returns a verdict of allow/block/escalate.
Audit Trail: Each action aligns with its corresponding Git file, providing a clear history of rule changes and approvals, which addresses the question, "Why did the agent make that decision?"

Setup Overview

Setting up gnt is quick and straightforward:

npm install -g @gnt-ai/cli
gnt login
gnt connect github
gnt prebrain
# merge the opened PR on GitHub for approval

This process generates structured rule files in your repository, resembling:

your-repo/
└── rules/
    ├── refund-approval-threshold.md
    └── contract-legal-cc.md

Rule Structure

Each rule file is written in plain Markdown with YAML frontmatter, showcasing key attributes such as:

---
title: Never refund over $500 without a manager
status: approved
confidence: 0.91
owner_id: finance-team
source_citations: [...]
source: slack
tags: [refunds, finance]
last_validated_at: 2026-07-20
version: 1
superseded_by: null
approved_by: jane@company.com
approved_at: 2026-07-21T14:03:00Z
created_at: 2026-07-18T09:12:00Z
pr_number: 142
pr_url: https://github.com/your-org/your-repo/pull/142
---
Refunds over $500 need manager sign-off before they go out...

Verification Response

The check_action tool evaluates an action before it occurs, returning structured JSON responses like:

{
  "verdict": "blocked",
  "reason": "Refund exceeds the $500 threshold without manager sign-off (rules/refund-approval-threshold.md)",
  "cited_rules": [
    { "id": "refund-approval-threshold", "title": "Never refund over $500 without a manager" }
  ],
  "rules_retrieved": 3
}

MCP Toolkit

gnt provides a suite of tools designed for comprehensive rule management:

Tool	Description
`check_action`	Validates actions against approved rules, returning verdicts with explanations.
`search_rules`	Performs semantic searches across approved rules with optional filtering.
`get_rule`	Retrieves a single approved rule by its ID, including its historical context.
`list_skill_packs`	Displays all compiled skill pack versions for your organization.
`get_skill_pack`	Accesses a skill pack's manifest and associated files by ID.

Privacy Considerations

Data Security: No analytics or telemetry are collected. Source texts processed in the cloud are sent directly to the API, with strict privacy protocols in place to protect sensitive information.
Self-Hosting: Organizations have the option to self-host gnt, providing full control over the infrastructure and rules management.

Learn More

For a deeper understanding of gnt and its capabilities, consult the Self-Hosting Walkthrough and the Security Policy.

Contribution

Contributions to gnt are welcome; refer to the Contributing Guidelines for more details on how to contribute effectively.