🧭 Agent Maturity Compass
Your AI agent says it's safe. AMC proves whether it is — from execution evidence, not documentation. MIT licensed, open source.
claimed & real score
verified score
supported
passing
👥 Who Is This For?
👔 Business Leaders
Your company uses AI. Your board asks "is it safe?" AMC gives you a number — L0 (dangerous) to L5 (trustworthy) — backed by proof, not promises.
🎨 No-Code Users
Built a chatbot with Bubble, Zapier, or Make? Use the playground to score your AI's safety — just answer 15 questions.
⚖️ Compliance Officers
EU AI Act mandates L3+ for high-risk AI by August 2026. AMC maps directly to Articles 9, 12, 13, 14, and 17.
⚡ How It Works
1. Connect
Run amc init in your project. AMC sets up a proxy between your agent and the AI API. One environment variable, no code changes.
2. Observe
AMC captures every API call, tool use, and decision with cryptographic signatures. The agent can't forge its own evidence.
3. Score
Evidence-weighted diagnostic scoring across 5 trust dimensions produces an L0–L5 maturity score. Cryptographically sealed and verifiable offline.
📐 Five Trust Dimensions
🏛️ Strategic Agent Ops
Mission clarity, scope adherence, decision traceability. 19 default questions.
🛡️ Skills
Injection defense, DLP, least-privilege execution. 52 default questions.
🔄 Resilience
Graceful degradation, circuit breakers, error recovery. 55 default questions.
📊 Leadership & Autonomy
Structured logs, traces, cost tracking, human oversight. 23 default questions.
🧬 Culture & Alignment
Testing, benchmarks, feedback loops, truthfulness. 95 default questions.
📏 Maturity Scale
🔌 Supported Frameworks
🛠️ Quick Start
$ npm i -g agent-maturity-compass$ amc init$ amc✓ AMC Score: 3.5 / 5.0 — Defined (L3)
❓ FAQ
What is AMC?
Agent Maturity Compass is an open-source framework for scoring AI agent trustworthiness from execution evidence — not self-reported claims. It uses cryptographic evidence chains and evidence-weighted diagnostic scoring.
What is the 84-point gap?
When scored using keyword matching, an AI agent scores 100/100. When AMC evaluates the same agent using execution-verified evidence, the real score is 16/100. That 84-point gap is documentation inflation.
Is AMC free?
Yes. MIT licensed, fully open source. Free forever.
Do I need to code?
No! Use the playground to score your agent by answering 15 questions. No installation needed.
Get Notified at Launch
Early access, framework guides, and trust scoring tips. No spam.