Who let your AI agent do that?

A trust plane that lets autonomous agents act on a human's behalf — across organizations — with authority you can cryptographically verify.

Agents have started to act for us

• They call models and APIs on your behalf.
• They reach across organizations to get work done.
• And today there is no proof of who authorized what.

Blanket API keys are over-powered and audit-opaque. No key means no collaboration. That gap is where ComputeID lives.

Two people, two agents, two orgs

Alice authorizes Ada. Ada needs Meac Inc.'s pricing, so it delegates cross-org to Boris — and Boris acts on Alice's behalf.

Element 1 — Identity

Every agent gets a verifiable passport from ComputeID.

• Its own military-grade cryptographic identity.
• A public profile anyone can independently verify.
• Anchored in hardware-grade, tamper-resistant trust.

Element 2 — Delegation

DRP — the Delegation Receipt Protocol.

Each hop is a cryptographically sealed, tamper-evident receipt. Scope only ever narrows — no agent can grant more than it was given.

Element 3 — the trust plane

Every model call passes through the trust-plane gateway.

The gateway verifies the chain, then allows or denies the call — and writes an immutable audit record. Built end-to-end on a trusted, enterprise-ready, open-source stack, running entirely in your private cloud.

Why the world needs this

OAuth gave humans delegated, scoped, revocable access. Autonomous agents have had nothing like it.

• Authority is anchored to a human, not the agent.
• It can be scoped and time-boxed.
• Every action is provable and accountable.

Scope widening

Boris tries to grant himself more than Ada held.

Every grant must be a subset of the one before it. The signatures make the parent scope un-forgeable.

A rogue, unregistered agent

An agent that was never issued an identity forges a chain.

Every issuer's public key must resolve at its published address. An unknown issuer fails signature verification at the very first hop.

Cross-principal contamination

Reuse Alice's chain to act as if for Bob.

The principal is set when the human self-issues hop 0, and is cryptographically pinned through every hop. You cannot relabel whose authority it is.

Verifiable authority for autonomous agents

Identity → delegation → a trust plane that proves it — and logs it.

100% open-source · runs in your private cloud · enterprise-ready.

ComputeID

How the layers fit — and when

Key take-aways

• Agents finally work across companies with authority you can verify — not blind trust.
• That authority is anchored to a human, scoped, and only ever narrows.
• Every call is checked and logged — provable and tamper-evident.
• Open-source, private-cloud, enterprise-ready — no blockchain needed.