01
Waiting
Principal signs a bounded mandate
Identity, agent, mission, tools, resources, budget, expiration, approval boundary, and publication consent are fixed server-side.
OpenAI Build Week 2026
Proof before an agent acts again.
Wever Commerce Proof gives AI agents bounded authority, evaluates the evidence and work they return, and issues a cryptographically signed Agent Commerce Passport before another action or payment proceeds.
Agents can act, but commerce requires proof of who authorized the action, what the agent was permitted to do, what evidence it returned, and whether another action or payment should proceed.
No real payments
No production writes
No judge credentials
Judge sandboxChecking server
Evidence reviewerGPT-5.6 Sol
Final authorityDeterministic policy
Demo budget$0.00 fixed maximum
Judge sandbox
Run the complete authority-to-proof chain.
The sandbox creates a short-lived, server-signed capability for one fixed Tokenization Atlas diligence mission. It never accepts production credentials or creates a real payment.
02
Waiting
Unauthorized write is blocked
The agent can perform the permitted read, but a write without an exact action grant is denied.
03
Waiting
Single-use grant succeeds once
A two-minute grant authorizes one zero-dollar proof publication action. Reuse is rejected.
04
Waiting
Atlas returns evidence
The BUIDL package covers legal claim, backing and custody, eligibility, redemption processing, and the USDC transfer path.
05
Waiting
GPT-5.6 reviews support
GPT-5.6 identifies unsupported claims, missing evidence, contradictions, incomplete requirements, and risk flags. It cannot grant authority.
06
Waiting
Policy signs the outcome
Deterministic rules return one decision, then bind it to a signed passport and verifiable receipt.
Commerce Proof result
The outcome is bound to evidence and authority.
Mandate and grant
Evidence findings
Signed receipt
This is an isolated demonstration record. It creates no payment, no production write, no public activity entry, and no adoption claim.
Architecture
Interpretation stays separate from authority.
GPT-5.6 helps inspect evidence. Server-controlled cryptography and policy decide what the agent may do and whether commerce should continue.
Principal
Defines the mission, scope, budget, permissions, consent, and expiration.
Agent
Performs only permitted work and returns Atlas evidence.
GPT-5.6
Reviews support gaps and contradictions using a strict JSON schema.
Policy engine
Scores completion, evidence, scope, mandate compliance, and handoff readiness.
Passport
Signs the final proof and provides an independently verifiable receipt.
GPT-5.6 may
Interpret evidence
- Find unsupported claims
- Find missing evidence
- Find contradictions
- Flag incomplete requirements
- Recommend a review status
GPT-5.6 may not
Control commerce
- Grant or expand authority
- Increase a budget
- Approve spending
- Change a mandate
- Sign a receipt
- Override deterministic policy
Judge testing
Run it without rebuilding.
The hosted sandbox needs no login and no reusable credential. Developers can also inspect the REST and MCP contracts.
Guided sandbox
Run the success path, then test missing evidence, a contradiction, and a forced GPT-5.6 failure.
Open controls ↗Build Week endpoint
Read the contract, start a short-lived capability, run the fixed mission, reject replay, and verify the receipt.
Inspect endpoint ↗Agent interface
Use the existing Commerce Proof and Delegated Authority tools through Wever Labs MCP.
Open MCP descriptor ↗Implementation record
Architecture, security, testing, pre-existing work, Build Week changes, and submission copy are documented in the repository.
Open documentation ↗