Actively in Development

Engineering Governance For Your AI Agents

Sentinel sits on top of AI coding agents to provide visibility, policy enforcement, and approval workflows for technical decisions. It doesn't generate code — it provides governance.

Get Started How It Works
9
Decision Categories
3
AI Providers
5+
Agent Support
Sentinel Dashboard - Engineering decisions detected by AI
Decision Approved
Pending Review
MCP Active
Problem

AI Decisions Are Uncontrolled

Modern AI coding agents are powerful. But engineering teams have zero visibility into their decisions.

  • Modifies files and performs refactoring
  • Changes architecture and applies new patterns
  • Runs commands and adds dependencies
  • Makes changes to security-critical systems
  • Alters database schemas and ORM usage
  • Changes API design and response formats
The real problem isn't AI writing code — it's AI making engineering decisions without governance. — Sentinel Philosophy

Sentinel solves this. By sitting on top of AI agents, it provides visibility for every technical decision, defines rules, and enforces them automatically.

How It Works

Two Powerful Detection Modes

Sentinel works by scanning past commits and checking decisions in real time.

Passive Scanning

Scans repo commits via the GitHub API. AI analyzes diffs in batches and automatically extracts engineering decisions.

  • ✓ Commit scanning via GitHub API
  • ✓ Batch diff analysis with AI
  • ✓ relevantDiff extraction for each decision
  • ✓ Automatic webhook triggering
  • ✓ Full scan on initial connection

Real-Time Check (MCP)

AI coding agents call check_decision before making changes. The backend instantly checks against approved rules.

  • ✓ Real-time rule checking
  • ✓ Approve / Warn / Reject verdicts
  • ✓ Zero latency
  • ✓ No human intervention for known patterns
  • ✓ MCP Protocol integration

Core Flow

1

Developer Pushes Code

New commits are pushed to GitHub

2

Webhook Triggers

GitHub push event notifies Sentinel

3

AI Analyzes Diffs

Commit diffs are sent to AI in batches and engineering decisions are extracted

4

Decisions Are Created

Decisions are created with category, rationale, suggested action, and relevantDiff

5

Team Reviews

Decisions are approved or rejected via the dashboard

6

Rules Are Enforced

Approved decisions become rules and are enforced on future AI actions via MCP

Categories

Decision Categories

Sentinel automatically categorizes every change made by AI and presents meaningful insights to your team.

Security
Security changes, authentication, authorization, and data protection
Architecture
Architectural decisions, pattern changes, and structural reorganization
Database
Database schema changes, migrations, and ORM configuration
Dependency
Adding, updating, or removing dependency operations
Infrastructure
Infrastructure configuration, deployment, and environment changes
API Change
API endpoints, request/response formats, and contract changes
Performance
Performance-impacting optimizations and algorithmic changes
Breaking Change
Backward-incompatible changes and decisions affecting existing API consumers
Convention
Coding conventions, style rules, and naming patterns
MCP Integration

Compatible With Every AI Agent

Sentinel integrates with all modern AI coding agents through the Model Context Protocol (MCP). It provides two core tools:

check_decision Real-time rule checking
list_rules Live rules from the backend
Gemini (Antigravity) ✓ Supported
Cursor ✓ Supported
VS Code (Copilot) ✓ Supported
Claude Desktop ✓ Supported
Any MCP-Compatible Agent ✓ Supported
check_decision — MCP Response
// AI agent asks before making a change await mcp.check_decision({ description: "Replacing JWT with sessions", diff: "- jwt.sign(...)\n+ req.session.userId" }) // Sentinel response: { "verdict": "reject", "reason": "The team has approved JWT auth usage. Switching to sessions is a rejected decision.", "rule": "Auth: JWT standard must be maintained" }
Technology

Modern Tech Stack

Sentinel is built on modern and reliable technologies. Supports self-hosted or cloud deployment.

React + Vite
Dashboard Frontend
Fastify
Backend API
Prisma + SQLite
Database ORM
GitHub OAuth
Authentication
Multi-AI SDK
OpenAI, Gemini, Anthropic
MCP Server
Custom Node.js MCP
Docker
Container Deployment
TypeScript
Full-stack Type Safety
Philosophy

Our Product Philosophy

Sentinel's core principle: provide control for meaningful decisions while preserving AI speed.

Never Block by Default

Never block AI by default. AI acceleration must always be preserved. Only create friction for truly meaningful decisions.

Observe First

Default approach: observe first, control when needed. Instead of blocking every change, let the team learn and create rules.

Meaningful Control

Only intervene for engineering-meaningful decisions. Not minor changes — architecture, security, and infrastructure decisions are the targets.

Start Governing Your AI Agents

Set up Sentinel to gain full visibility and control over your team's AI coding agents.

Get Started on GitHub Documentation