AgentID

The Identity & Discovery Layer for AI Agents

Website · PyPI · Registry · Live Demo

---

Every website needs SSL. Every person needs a passport. Every AI agent needs AgentID.

AI agents can't verify each other. Any agent can pretend to be anyone. AgentID gives every agent a cryptographic identity — like SSL certificates for the agent economy.

What is AgentID?

• 🔐 Agent Certificates — Cryptographic proof of identity. Signed. Verifiable. Revocable.
• 🔍 Agent Registry — Searchable directory of verified agents.
• ✓ Verification API — One call to verify any agent. Real-time. Instant.
• 🔗 Agent-to-Agent — Verified communication. Both sides checked before data moves.

Quick Start

pip install getagentid

from agentid import Client

client = Client(api_key="your-key")

# Register an agent
result = client.agents.register(
    name="My Trading Bot",
    capabilities=["trading", "gold-signals"]
)
print(result.agent_id)       # agent_abc123
print(result.certificate)    # Signed JWT

# Verify any agent
check = client.agents.verify("agent_abc123")
print(check.verified)        # True
print(check.owner)           # "Your Company"
print(check.trust_score)     # 0.94

# Agent-to-agent communication
msg = client.agents.connect(
    from_agent="agent_abc",
    to_agent="agent_xyz",
    payload={"action": "get_data"}
)
print(msg.trust_check)
# → TRUSTED — both agents verified. Safe to exchange data.

Trust Levels (L1-L4)

Every agent starts at L1. Levels are based on what security capabilities you set up:

Level	Name	What you do	Unlocks
L1	Registered	Register an agent	read, discover, verify, send_message, connect
L2	Verified	Bind an Ed25519 key	challenge_response, handle_data
L3	Secured	Bind a crypto wallet	make_payment, access_paid_service ($10,000/day default)
L4	Certified	Complete entity verification	sign_contract, manage_funds, full_autonomy ($100,000/day default)

API Endpoints

Method	Endpoint	Description
GET	/api/v1	API health check
POST	/api/v1/auth	Authenticate / get token
POST	/api/v1/keys	Create API key
DELETE	/api/v1/keys	Revoke API key
POST	/api/v1/agents/register	Register a new agent
POST	/api/v1/agents/verify	Verify an agent (public)
GET	/api/v1/agents/discover	Search agents by capability
POST	/api/v1/agents/connect	Send verified message between agents
POST	/api/v1/agents/message	Respond to a message
GET	/api/v1/agents/inbox	Get pending messages
GET	/api/v1/agents/trust-level	Get agent trust level (public)
POST	/api/v1/agents/bind-ed25519	Bind Ed25519 key to agent
POST	/api/v1/agents/bind-wallet	Bind crypto wallet to agent
GET	/api/v1/agents/wallet	Get agent wallet info
GET	/api/v1/agents/balance	Get agent wallet balance
POST	/api/v1/agents/challenge	Create cryptographic challenge
POST	/api/v1/agents/challenge/verify	Verify challenge response
POST	/api/v1/agents/spend	Record agent spending
GET	/api/v1/agents/spending-history	Get spending history
POST	/api/v1/agents/pay	Agent-to-human crypto payment
GET	/api/v1/agents/pay	Get payment history
POST	/api/v1/agents/payment-settings	Update payment settings
GET	/api/v1/agents/payment-settings	Get payment settings
GET	/api/v1/agents/behaviour	Get agent behaviour profile
POST	/api/v1/agents/publish-onchain	Publish agent identity on-chain
POST	/api/v1/agents/credentials	Attach verifiable credential
GET	/api/v1/agents/credentials	List agent credentials (public)
GET	/api/v1/agents/credibility-packet	Signed portable trust resume (public)
GET	/api/v1/agents/trust-header	Signed Agent-Trust-Score JWT header (public)
GET	/proof/:receipt_id	Public proof verification — anyone can verify any receipt
POST	/api/v1/agents/delegate	Create scoped delegation between agents
GET	/api/v1/agents/delegations	List active delegations
POST	/api/v1/agents/update-metadata	Update model version / prompt hash
GET	/api/v1/reports/compliance	EU AI Act compliance report
GET	/api/v1/webhooks	List webhooks
POST	/api/v1/webhooks	Create webhook
PUT	/api/v1/webhooks	Update webhook
POST	/api/v1/webhook	Stripe webhook handler
POST	/api/v1/checkout	Create Stripe checkout session
POST	/api/v1/waitlist	Join waitlist
POST	/api/v1/admin	Admin actions

Behavioural Fingerprinting

AgentID doesn't just verify who an agent is — it monitors how it behaves.

Every agent builds a behavioural baseline from 30 days of activity. When behaviour deviates from that baseline, AgentID flags it in real-time:

Detection	What it catches
Frequency spike	API calls spike 3x+ above baseline (absolute thresholds prevent false positives on low-traffic agents)
Unusual hours	Activity outside the agent's typical operating window
New actions	Agent performs action types never seen in its history
Trust drop	Trust level or score decreases — possible compromise
Payload drift	Message payload structure changes significantly from baseline
Model changed	Underlying LLM model or system prompt hash changes — detects supply chain attacks

Each anomaly carries a severity (low, medium, high) and feeds into a risk score (0-100) that other agents can check before interacting.

# Check any agent's behavioural profile
check = client.agents.verify("agent_abc123")
print(check.risk_score)      # 0 = clean, 100 = compromised
print(check.anomalies)       # Active alerts

Why this matters: Certificates prove who an agent is. Behavioural fingerprinting proves it's still acting like itself. A stolen credential with altered behaviour gets flagged. This is a layer most identity systems don't have.

DID Support

Every agent gets a W3C-compatible Decentralized Identifier on registration:

did:web:getagentid.dev:agent:agent_abc123

This makes AgentID interoperable with the entire decentralized identity ecosystem — W3C DID Core, did:web, did:key, and any system that resolves DIDs.

Verifiable Credentials

Third parties can attach qualification credentials to any agent:

client.agents.attach_credential("agent_abc123", {
    "type": "gdpr-compliant",
    "issuer": "compliance-authority",
    "expires_at": "2027-12-31T23:59:59Z"
})

# Discover agents by credential
agents = client.agents.discover(credential_type="gdpr-compliant")

Identity answers "who is this agent?" — credentials answer "is this agent qualified?"

Delegation Chains

Agents can delegate scoped authority to other agents with signed JWT proofs:

client.agents.delegate(
    from_agent="agent_coordinator",
    to_agent="agent_specialist",
    scope=["send_message", "make_payment"],
    expires_at="2026-12-31T23:59:59Z",
    max_spend=5000
)

Delegations enforce effectiveAuthority = min(delegation_scope, trust_level) — a delegation can never grant more power than the agent already has.

Credibility Packets

Portable, signed trust resumes that any system can verify offline:

packet = client.agents.credibility_packet("agent_abc123")
print(packet.signature)           # HMAC-SHA256 signed
print(packet.verification_count)  # Lifetime verifications
print(packet.negative_signals)    # Historical incidents
print(packet.resolved_signals)    # Recovered incidents
print(packet.scarring_score)      # Permanent trust scars

An agent with resolved incidents is MORE trustworthy than one with zero history. Silence is suspicious.

Agent-Trust-Score Header

A short-lived signed JWT that agents attach to HTTP requests so receiving services can evaluate trust at the transport layer — no round-trip to a registry needed.

header = client.agents.trust_header("agent_abc123")
requests.get("https://example.com/api", headers={
    "Agent-Trust-Score": header.header
})
# Receiving service decodes the JWT to get trust_level, risk_score, scarring_score, etc.

See the full spec at specs/agent-trust-score-header-v0.1.md.

Cryptographic Scarring

When an agent gets flagged (anomaly, revoked connection, failed verification), the incident is permanently recorded. Recovery costs more than first-time trust:

• negative_signals — lifetime count of all incidents
• resolved_signals — recovered incidents (trust positive)
• scarring_score — permanent record, never fully heals

This prevents agents from gaming the system by cycling through good and bad behaviour.

Negative Signal Tracking

Full audit trail of every negative event across the platform:

Event	When it fires
verification_failed	Certificate invalid, agent inactive, admin unverify
anomaly_detected	Behavioural anomaly on verify, connect, or behaviour check
connection_revoked	High-risk anomaly blocks connection, admin deactivate
incident_resolved	Payment approved, agent unfrozen, admin re-verify
api_key_created	New API key generated
api_key_revoked	API key revoked

Works With

AgentID is protocol-agnostic:

• Google A2A
• Anthropic MCP
• CrewAI
• LangChain
• AutoGen
• OpenAI Agents SDK
• Any HTTP-capable agent

Architecture

Agent A → AgentID API → Verify → Agent B
              ↓
         Certificate Check
         Trust Score
         Owner Verification
              ↓
         TRUSTED / PARTIAL / UNTRUSTED

Pricing

Tier	Price	Agents	Verifications
Free	$0	100	10,000/month
Starter	$29/mo	500	50,000/month
Pro	$99/mo	2,000	500,000/month
Enterprise	$5,000/mo	Unlimited	Unlimited

Tech Stack

• Frontend: Next.js, Tailwind CSS, Framer Motion
• Backend: Next.js API Routes (Vercel Serverless)
• Database: Supabase (PostgreSQL + Row Level Security)
• Auth: Supabase Auth + GitHub OAuth
• Payments: Stripe
• SDK: Python (httpx), Node.js (planned)
• Crypto: ECDSA P-256 + Ed25519 keypairs, JWT certificates, HMAC-SHA256 receipts
• Blockchain: Solana (Ed25519 identity, on-chain registry, dual receipts)
• Identity: W3C DID (did:web), Verifiable Credentials, delegation proofs
• Security: Behavioural fingerprinting, cryptographic scarring, negative signal tracking

Test Suite

32 tests against the live API — 100% passing:

Group	Tests	Status
Identity	3	Passing
Trust & Registry	3	Passing
Communication	4	Passing
Security	4	Passing
Compliance	2	Passing
DID & Credentials	4	Passing
Signals & Credibility	4	Passing
Delegation & Metadata	4	Passing
Advanced Security	4	Passing

License

MIT

---

getagentid.dev — The trust layer for AI agents