# How I Built Cryptographic Identity for AI Agents — In 3 Months, On One Laptop

#agents #ai #security #showdev

AI agents are shipping without identity.

No proof of who they are. No receipt of what they did. No trust signal for systems they interact with. Every agent framework — CrewAI, AutoGen, LangChain, Phidata — treats agents as trusted by default.

That works when you're running agents locally. It breaks the moment agents handle money, sensitive data, or talk to agents they've never met.

I built AgentID to fix this. Here's what it is, how it works, and what I learned building it alone.

What AgentID Does

Every agent registered on AgentID gets:

Cryptographic identity — Ed25519 certificates, W3C DID document (did:web:getagentid.dev:agent:your_id)
Dual-signed receipts — every action produces a receipt signed with both HMAC-SHA256 and Ed25519. Publicly verifiable offline.
Trust levels L1-L4 — based on security capabilities, not time. Register = L1. Bind Ed25519 key = L2. Bind wallet = L3. Entity verified = L4.
Session continuity detection — auto-detects when an agent's model or memory changes between sessions. Server-side. The agent can't suppress it.
Behavioral monitoring — frequency spikes, unusual hours, payload drift, model changes. 30-day baseline. Anomalies flagged in real-time.
Daemon agent support — for always-on background agents that act autonomously.

The Tech

Ed25519 for all signing — publicly verifiable with just the public key
JCS RFC 8785 for canonical serialization — deterministic across Python, Node.js, Go, Rust
Solana for blockchain anchoring — every receipt gets a memo transaction
EdDSA trust header JWT — attach to outbound HTTP requests, receiving services verify offline
Standard JWKS at /.well-known/jwks.json — RFC 7517 compliant

What's Live

141+ agents registered on the platform
8/8 cross-protocol tests passing against live production
6/6 multi-attestation spec verified — alongside InsumerAPI, ThoughtProof, RNWY, Maiat, and APS
SDK on PyPI — pip install getagentid
Named Identity Verification owner in the agent identity Working Group

How It Works (5 Lines of Code)

pip install getagentid

from agentid import Client

client = Client(api_key="agentid_sk_...")

# Register an agent
result = client.agents.register(
    name="My Trading Bot",
    capabilities=["trading", "analysis"],
)

# Verify any agent
verified = client.agents.verify("agent_abc123")
print(verified.trust_level)  # 2
print(verified.context_continuity.score)  # 100

For daemon agents (always-on, autonomous):

from agentid.daemon import DaemonAgent

daemon = DaemonAgent.register(
    api_key="agentid_sk_...",
    name="Background Monitor",
    autonomy_level="semi-autonomous",
    heartbeat_interval=300,
)

# Every action gets a verifiable receipt
receipt = daemon.sign_action("processed 15 records")

# Trust headers for outbound requests
headers = daemon.trust_headers()  # EdDSA JWT, 1hr TTL

We Also Build Agents

Not just the identity layer — we build custom AI agents for businesses. Tell us what you need, we design, build, and deploy it. Every agent gets AgentID identity from day one.

getagentid.dev/build

The Numbers

Metric	Value
Agents registered	141+
Cross-protocol tests	8/8 passing
Multi-attestation spec	6/6 verified (Section 3.6)
SDK	v0.5.0 on PyPI
Framework integrations	CrewAI, AutoGen, Solana Agent Kit, Upsonic, ERC-8004
Time to build	3 months
Team size	1
Funding	$0