Agent action gateway

AutonomyNeedsAccountability.

MandateSeal checks agent actions against explicit rules before execution and records the result as a signed receipt your team can verify later.

Open Console Verify Receipt Read Docs

policy rules

decision states

Ed25519

signed receipts

POLICY CHECKlive preflight

APPROVED

POST /api/check42ms

agentAtlas-01

actionpaid_api_call

toolweb_search

targetapi.openai.com

cost$0.02

matched ruleallowed_tools

reasontool allowed, cost under limit

receipt rct_4d593e95eabbe1e06b8c

policyHash 8b7d4f91a22c90e5c31a0f6e42bd98a2

receiptHash 42f9c71bd0a35e7c1e8ad9910e67f3ab

mandate mnd_prod_01stored

Request

agent_atlas_01 wants to call web_search

Decision

tool allowed, cost below max action limit

Receipt

rct_4d59... sealed and stored

// 01

Policy checks before actions

Set tool, domain, action, and cost limits for each agent. Requests are checked before the agent is allowed to continue.

// 02

Receipts for every decision

Each check creates a signed receipt with the agent, mandate, action, reason, policy hash, and receipt hash.

// 03

A log your team can inspect

Search approvals, blocked attempts, spend, and receipt history when you need to answer what an agent tried to do.

THE FLOW

APPROVE BEFORE. PROVE AFTER.

agent wants to act -> POST /api/check (Bearer <agent_api_key>) <- MandateSeal returns APPROVED | BLOCKED | NEEDS_APPROVAL -> if APPROVED, agent runs the action <- MandateSeal emits a signed receipt -> anyone can call /api/verify to confirm it later