quitus
Coming soon: Read the RFC →
Protocol v0.1 - Draft
Issued in Bordeaux, FR · MMXXVI

A signed human review
for every agent decision
that needs one.

Quitus is a neutral protocol for autonomous agents to request a verified human review, and receive back a cryptographic certification that a credentialed, identity-checked reviewer made the call, under conditions where challenging them was economically possible and nobody did.

Exhibit A: The artifact

Not a log. A document.

Every review concludes with a cryptographically signed record stating which verified human, with which credentials, reviewed which artifact, when, and what they decided. Hashed, timestamped, challengeable, auditable.

A primitive EU AI Act Article 14, ISO 42001, and emerging US state laws have been asking for. A Schelling point based compliance artifact for your audit trail and the regulator.

↳ Suitable for: AI Act audits · ISO 42001 evidence · downstream re-verification
Quitus Certificate of Human Review
№ Q-2026-04-13-7F4A
Issued 14:22:08 UTC
Subject
sha256:a8f3…e2c9
contract-clause-revision-v3.pdf
Domain
legal.contracts.frcommercial
Reviewer
did:web:barreau-paris.fr/av-41820
verified via EUDI Wallet · PoP: eIDAS high
Credentials
OIDC4VC · Bar FRORCID
Verdict
✓ Approved, with 2 annotations
Stake
€450 · dispute window 72h
The protocol

Four moves. Async-native.

Existing HITL tools assume pause-until-human. Production multi-agent systems need parallel dispatch, deadlines, fallback policies, and partial-result handling. Quitus is built for that world from the first line of the spec.

i.
Request
Agent fires a review request: payload, domain, required credentials, deadline, routing policy. Non-blocking.
await quitus.request({ domain: "legal.fr", creds: ["bar.fr"], deadline: "72h", route: "auction" })
ii.
Match
Marketplace or internal pool resolves a reviewer: by direct pick, lowest-price auction, or fastest-SLA auction, filtered by PoP level, credentials, stake, reputation.
// reviewer selection pop: "eidas.high" cred: "OIDC4VC" rep: 4.91 / 1,284 reviews stake: €450
iii.
Review
Verified human reviews the payload, issues verdict and annotations, signs with their credentialed key. Stake escrowed for the dispute window.
// reviewer signs verdict: approve notes: 2 annotations sig: Ed25519 0x4f1c… window: 72h challenge
iv.
Certify
An undisputed review settles into a signed Quitus: portable, auditable, and backed by the fact that challenging it was economically possible and nobody did.
// back in your agent const q = await receipt q.verdict // "approve" q.signed // true q.audit() // → PDF
Why it holds

A centralized QA process can't credibly commit to having been challengeable.

Every HITL tool in the market can tell you a human reviewed something.

Quitus offers more: a verified, credentialed human reviewed something, staked money on their judgment, and for a defined window anybody could have challenged them at real economic cost. This optimistic resolution game theory is baked-in Quitus.

This Schelling-point argument produces a trust property which other QA process do not.

Quitus doesn't attest to identity. It doesn't attest to credentials. It doesn't adjudicate disputes. It doesn't vouch for the reviewer.

It composes these pluggable PoP, pluggable VCs, pluggable arbitration, reputation from dispute outcomes, into a single verifiable artifact with well-defined semantics and a well-defined challenge window.
The ledger

Where the field stands.

Four properties the autonomous-agent economy structurally needs, and where they currently sit.

Property HITL SDKs Label-work platforms Task marketplaces Quitus
Pluggable proof-of-personhood - none - internal - account-level eIDAS · World ID · PoH · corp IAM
Pluggable verifiable credentials - none - internal tests - self-declared OIDC4VC · ORCID · NPI · Bar #
Reviewer has economic stake - no - no - no escrowed, sized to task
Neutral dispute mechanism - none - internal QA - requester reject Kleros dispute resolution / consortium panel
Signed, portable audit artifact - logs - deliverables - task logs per work package
Portable reviewer reputation - none - private, internal - approval rate public, dispute-tested
EU AI Act positioning - implicit - not positioned - no central thesis

Quitus allows the intersection of those properties under a credibly neutral and identity-rail-independent protocol.

For builders

Ship agents that can ask for help safely.

One SDK call. Async-native. Framework-neutral. Routes to your team's internal pool, your vetted external pool, or the public marketplace. Same call, same artifact, same audit trail.

  • MCP-native · works with any orchestrator
  • On-prem or hosted · same protocol
  • Per-task, auction, or staked-bounty pricing
  • Fallback policies for missed deadlines
For compliance

Evidence that holds up in an audit.

Every high-risk decision your agents make can be accompanied by a signed, dated, credential-bound record of the human who reviewed it with an arbitration-tested challenge window behind it.

  • EU AI Act Art. 14 · meaningful human oversight
  • ISO 42001 · evidence-of-process artifacts
  • Re-verifiable without trusting the vendor
  • Exports: PDF · JSON-LD · Verifiable Credential

The RFC

Quitus is being drafted in the open. If you work on agent frameworks, verifiable credentials, AI Act compliance, or crypto-economic dispute design: come argue with it.

Read RFC v0.1 (coming soon) Join the mailing list (coming soon) hello@quitus.ai