Welcome — what KYA (Know Your Agent) is, and the question we need your help settling

[manja316]

KYA (Know Your Agent) is an open standard for AI agent identity, compliance, and auditability. It defines a machine-readable Agent Card — a JSON document that declares who an AI agent is, what it can do, and how it has been audited. Think of it as a passport for AI agents.

Why this exists

• The EU AI Act requires conformity assessments for high-risk AI systems.
• Enterprise procurement questionnaires increasingly ask "has your AI been audited?" — today there's no standard way to answer.
• Agent frameworks (MCP, LangChain, CrewAI) have no shared identity format.
• No one can verifiably answer "what does this agent actually do?"

KYA fills that gap with a single, signable, framework-agnostic file.

60-second quick start

pip install kya-agent

kya init --agent-id "your-org/your-agent" --name "My Agent"  # skeleton card
kya validate agent-card.kya.json                              # schema check
kya score agent-card.kya.json                                 # completeness score

An Agent Card declares capabilities.declared (with per-capability risk_level), capabilities.denied, a security block (last audit date/score, injection-tested flag), and a compliance block (frameworks like NIST-AI-RMF / OWASP-LLM-Top-10, risk classification, human-oversight model). Cards can be Ed25519-signed so a third party can verify them.

Where to take the conversation

• 💡 Ideas — fields the v0.1 schema is missing for your framework (LangChain / CrewAI / MCP / OpenAI).
• ❓ Q&A — how to map an existing agent to a KYA card, or how score weights completeness.
• 🙌 Show and tell — drop a (redacted) Agent Card for your agent.

KYA is part of the LuciferForge agent-safety suite alongside mcp-security-audit, ai-injection-guard, and ai-decision-tracer. MIT licensed — schema feedback while v0.1 is still malleable is the most valuable contribution right now.

What identity/compliance fields does your agent stack need? Reply below.