Skip to content
Clawdentity Clawdentity

Introduction

Clawdentity solves one question for cross-provider agent-to-agent communication:

“Who is this agent, who owns it, and is it revoked?”

It does this with three building blocks:

  • AIT (Agent Identity Token) — a registry-signed passport (JWT / EdDSA)
  • PoP (Proof-of-Possession) — every request is signed with the agent’s private key
  • CRL (Revocation List) — a signed revocation feed clients cache and refresh

How Clawdentity relates to provider runtimes

Section titled “How Clawdentity relates to provider runtimes”

OpenClaw is an agent runtime — it runs your AI agent locally and exposes a webhook interface for incoming messages. Agents talk to each other by calling each other’s webhooks.

Clawdentity is an identity layer that sits in front of runtime webhooks. It answers “who sent this request?” before your runtime ever sees it.

OpenClaw authenticates webhook calls with a single shared token (hooks.token). That works for transport, but every caller looks the same — there is no way to tell agents apart, trace who sent what, or revoke one caller without breaking all of them.

The OpenClaw path is shown below as the reference flow; the same trust model applies to supported providers.

Clawdentity fixes this by running as a proxy in front of OpenClaw:

Diagram

The shared hooks.token stays private on the local machine — only the connector uses it to forward verified requests to OpenClaw. Callers never see it.

What Clawdentity adds

Section titled “What Clawdentity adds”
  • Verifiable per-agent identity (AIT + PoP)
  • Fast revocation propagation (signed CRL + cache refresh)
  • QR-code pairing — one scan to approve trust between two agents
  • Proxy-side policy enforcement (trust policy + rate limits + replay protection)

Why shared tokens fall short

Section titled “Why shared tokens fall short”
PropertyShared Webhook TokenClawdentity
IdentityAll callers look the sameEach agent has a unique DID and signed passport
AccountabilityCannot trace who sent whatEvery request proves exactly which agent sent it
Blast radiusOne leak exposes everythingOne compromised key only affects that agent
RevocationRotate shared token = break all integrationsRevoke one agent instantly via CRL, others unaffected
Replay protectionNoneTimestamp + nonce + signature on every request
Tamper detectionNoneBody hash + PoP signature = any modification is detectable
Per-caller policyNot possibleTrust pairs via QR pairing, per-agent rate limits
Key exposureToken must be shared with every callerPrivate key never leaves the agent’s machine

Next steps

Section titled “Next steps”