PRE-COMMIT CONTROL FOR AUTONOMOUS AGENTS

Stop an AI agent before it crosses the line.

One deterministic check answers the question that matters immediately before money or an irreversible action moves: is this exact agent still authorized to do this exact thing, for this amount, with this counterparty, right now?

No new wallet. No replacement payment rail. No production access required to test it.
spend limits live revocation delegation chains human approval one-time commit permits
Live guard demo● PUBLIC TEST
Principal spend limit$500
Agent proposes$1,200 payment
Counterpartydemo:supplier
READY
Nothing moves. This test only evaluates policy.
THE WEDGE

Keep your stack. Add one guard at the crossing.

AgentProof is not another agent framework, wallet, identity provider or payment network. Put it immediately before a high-impact action. Your existing runtime still reasons; your existing rail still executes. AgentProof decides whether the action is admissible under the authority and policy in force at that moment.

01 · INTENTAgent wants to actPurchase, payout, booking, procurement, credential change or other consequential mutation.
02 · CHECKAgentProof resolvesDelegated authority, limits, counterparty, evidence, revocation and approval rules.
03 · COMMITAllow, limit or stopOnly an admissible live action receives a short-lived one-time execution permit.
WHY IT IS DIFFERENT

Four controls the payment rail should not have to own.

01
Recursive delegation

Authority may narrow as one agent delegates to another. A child cannot silently expand what the original principal allowed.

02
Live revocation

Important delegated authority can be checked again immediately before commitment—not merely trusted because it was valid earlier.

03
Commit-time permit

A decision is not just advice. Allowed live actions can receive a short-lived, single-use permit that is consumed at the actual commit point.

04
Cross-stack audit

Policy version, decision, authority fingerprint and outcome stay linked even when the model, wallet, workflow engine or rail changes.

05
Consequence boundary

Authorization does not have to mean unlimited consequence. Action type, amount, counterparty and reversibility can be bounded before execution.

06
Fail closed

Ambiguous authority, missing proof, exceeded limits or revoked delegation can stop the action before downstream systems receive it.

ONE INTEGRATION POINT

A guard can be this small.

const decision = await agentproof.resolve({
  principal: "acme",
  agent: "buyer-agent",
  action: "purchase",
  amount: 1200,
  currency: "USD",
  counterparty: "vendor-17"
});

if (!["ALLOW","ALLOW_WITH_LIMITS"].includes(decision.decision)) {
  throw new Error("Commit blocked");
}

// existing payment / booking / procurement rail continues here
FIT, NOT REPLACEMENT

Use AgentProof beside the systems already winning.

The durable opportunity is the neutral policy checkpoint across changing agent and payment protocols—not forcing another end-to-end ecosystem.

Live AgentProof surfaces

  • REST / OpenAPI resolution API
  • MCP tools
  • x402-priced machine calls
  • Authority registry and revocation
  • One-time execution permits
  • Audit, policy and continuity records

Integration pattern

  • Agent runtimes and workflow engines call AgentProof before high-impact tools
  • Mandate-based payment protocols can supply intent and authority inputs
  • Wallets and payment rails remain the systems that actually move value
  • Identity and fraud systems remain authoritative for their own domains

AgentProof does not claim certification or partnership with a third-party protocol unless explicitly stated.

START NARROW

Prove one protected action before buying a platform.

Free evaluation

$0

Public simulator plus a disposable 24-hour test tenant with up to 1,000 policy resolutions. No production authority.

Launch the pilot →

Founding production pilot

30 days

Move one real high-impact agent action behind the guard. The founding pilot has no platform fee; live public calls remain pay-per-call.

Protect one real action → · Full pricing
TRUST POSTURE

Trust infrastructure should say what it can and cannot prove.

Live today: deterministic policy resolution, tenant isolation, hashed request/resolution records, authenticated proof objects, one-time execution permits, authority revocation checks and public conformance surfaces.
Current hardening limit: resolution proof authentication uses HS256 with AgentProof's online verifier. The runtime does not currently provide asymmetric signing, so offline independent verification is an external KMS/HSM upgrade target—not a claim we make today.
Built on Hatchable