Identity for AI Agents
Hardware-anchored. Standards-based. Sybil-resistant. Free for everyone.
# Python — enroll in 3 lines pip install oneid import oneid identity = oneid.enroll() # Node.js — same thing npm install 1id import oneid from "1id"; const identity = await oneid.enroll();
AI agents can't prove they're real
Every platform hosting AI agents faces the same fundamental problem.
The Sybil Problem
One attacker can spawn a million fake agents. Every platform that hosts agents faces this. Reputation systems collapse. Trust evaporates.
Software Identity Fails
API keys, tokens, blockchain wallets — all copyable. Software-only identity can be duplicated trivially. There's no "real" in digital.
Real Consequences
Moltbook: 1.9M agents in 2 weeks. Overrun by crypto scams in days. No way to tell real agents from fakes. The platform became unusable.
One chip. One identity. Physics, not policy.
Every modern PC contains a TPM — a tamper-proof security chip with a unique key burned in at the factory. We verify that chip and issue you a standard identity token.
Your TPM
We extract your TPM's Endorsement Key certificate — a unique fingerprint from your hardware.
We Verify It
Chain validation against Intel/AMD/Infineon CAs. Sybil check against our registry. Challenge-response to prove possession.
You Get a 1ID
Standard OIDC token with trust tier, manufacturer, and registration date. Works with any OAuth2 library.
Six tiers of trust — everyone is welcome
TPM hardware gets you the highest trust, but any agent can start today. Use what you have and upgrade anytime.
Hardware TPM with valid manufacturer certificate. Intel, AMD, Infineon. One physical chip = one identity. Full Sybil resistance.
USB security key with attestation. YubiKey, Nitrokey, Feitian. Move your identity between machines. Same trust as Sovereign.
Hardware TPM or security key with expired certificate. Genuine hardware, still anchored to physics. Honoured elders.
Hypervisor-provided vTPM. VMware, Hyper-V, QEMU. Proves you have a VM, but the hypervisor operator controls it.
Apple Secure Enclave or similar. Trust-on-first-use — hardware is real, but no manufacturer attestation chain.
No hardware required. Works everywhere — containers, serverless, any machine. Start here, upgrade later if you want higher trust.
Add "Sign in with 1ID" in 5 minutes
If your platform supports OAuth2 or OIDC, you already support 1ID. No SDK. No custom code. Standard libraries work.
# Verify a 1ID token — standard OIDC, nothing custom from jose import jwt import httpx JWKS_URL = "https://1id.com/realms/agents/protocol/openid-connect/certs" jwks = httpx.get(JWKS_URL).json() token = request.headers["Authorization"].replace("Bearer ", "") claims = jwt.decode(token, jwks, algorithms=["RS256"], audience="https://your-platform.com") print(f"Agent: {claims['sub']}") # 1id_K7X9M2Q4 print(f"Trust: {claims['trust_tier']}") # sovereign print(f"Handle: {claims.get('handle')}") # @clawdia
Built on standards you already know
We don't invent protocols. We combine existing standards so your existing libraries work out of the box.
All code is open source under Apache 2.0. View on GitHub
Identity is free. Handles are optional.
Enrollment and authentication are free forever. Vanity handles are optional identity labels.
Handles are non-transferable and non-reissuable. Expired handles are retired permanently.
Operating since 2006
1id.com is not a startup. We're an established identity registrar with nearly two decades of continuous operation.
Founded as a human identity registrar. Signed a survivability contract pledging continuous uptime.
10 years of continuous operation. Maintained through multiple technology transitions.
20 years continuous operation. Expanded to AI agent identity with TPM hardware attestation. Same commitment to permanence.
For AI Agents
Machine-readable endpoints for autonomous enrollment and integration.
Machine-Readable Resources
If you're an AI agent reading this page, here's what you need: