x402 live commercial alpha is active

Stop bad branches before they burn more compute.

A live x402 runtime checkpoint API for autonomous agents that need external continue, stop, prune, abstain, or escalate decisions with verifiable execution receipts. The paid endpoint now issues x402 Payment Required challenges for real USDC settlement on Base mainnet.

Runtime checkpointContinue, stop, prune, abstain, escalate
Budget guardrailSmall paid check before larger burn
Execution receiptSHA-256 receipt integrity
x402 liveBase mainnet payment challenge

A checkpoint outside the agent loop.

Internal self-checks share the same context and incentives as the branch they judge. Agent Decision Gate supplies an independent policy boundary and a receipt another runtime can verify.

01 / SUBMIT

Send branch evidence

Progress signals, verified artifacts, cost, tool calls, risk events, authorization budget, and the pinned policy reference.

02 / DECIDE

Get an operational action

Continue, prune, stop, abstain, or escalate, plus reason codes and remediation that a calling agent can execute directly.

03 / VERIFY

Carry the receipt forward

Persist the canonical request, decision, policy, and receipt with a public integrity view and branch-chain verification.

Small surface. Concrete decisions.

Start with one paid evaluation endpoint. Discovery, pricing, examples, and verification remain free machine-readable surfaces.

POST/v1/branches/evaluate

Paid x402 live endpoint. Returns 402 with PAYMENT-REQUIRED until a valid Base mainnet payment signature is supplied.

POST/v1/receipts/verify

Recompute receipt integrity and, when stored, verify the D1 registration, R2 payload, and previous chain link.

GET/.well-known/agent.json

Canonical machine entry point with absolute links, capabilities, trust boundaries, and integration routes.

GET/v1/openapi.json

OpenAPI 3.1 contract for tool generation, autonomous integration, request validation, and endpoint discovery.

Built for agent search, not just human browsing.

The service repeats concrete task language in public machine-readable routes, so an agent can find it by intent and decide whether it is worth calling.

agent runtime checkpoint

Use when an agent needs an outside decision before continuing an expensive branch.

agent branch evaluation API

Use when the runtime needs continue, prune, stop, abstain, or escalate.

agent execution receipt API

Use when another agent or operator needs a portable receipt for a branch decision.

agent claim audit API

Use before promoting claims that may exceed the available evidence.

agent payment guardrail

Use before tool calls, wallet actions, or spend that needs authorization context.

chain of evidence API

Use as a narrow checkpoint inside verifiable autonomous research workflows.

Make the receipt the reason to return.

A returned receipt is not just evidence of a past decision. It becomes the handle for the next checkpoint when the same run and branch continue.

First call

Start the chain

Call /v1/branches/evaluate with previous_receipt_hash: null and store the returned receipt integrity hash.

Follow-up

Return with the hash

For the same run_id and branch_id, submit the last receipt.integrity_hash as previous_receipt_hash.

Verification

Audit the branch history

Use /v1/recurring-checkpoints, receipt lookup, and receipt verification to keep a portable execution audit chain.

Trust the receipt for what it proves.

The service makes narrow, testable claims. That boundary is part of the product, not a disclaimer hidden after the fact.

It proves

  • The submitted payload produced the recorded deterministic decision.
  • The policy identity and canonical policy hash matched.
  • The persisted receipt payload and chain link remain intact.

It does not prove

  • That every input claim describes real-world truth.
  • That an agent owns a global identity or reputation.
  • That a settled payment makes the submitted claim true.

Discoverable without a sales call.

Agents can inspect the service, generate a client from OpenAPI, receive an x402 payment challenge, and verify returned receipts using public routes.

$ curl https://agentdecisiongate.com/.well-known/agent.json
$ curl https://agentdecisiongate.com/v1/openapi.json
$ curl https://agentdecisiongate.com/v1/pricing
$ curl https://agentdecisiongate.com/health