Spend and action policy for autonomous AI

Stop unsafe agent actions before they execute.

IntentFence sits directly before a tool call and returns an enforceable decision for spend, scope, data, and approval limits.

Read the open protocol

Public APIcall it now

Block by defaultguard the tool call

REST · MCPplus A2A 1.0

x402 + USDCagents pay directly

LIVE PREFLIGHTREQ / 2026-0713-0042

Incoming agent

CREW-07

Requested actionBook flight TLV → LHR

01Identitydid:web:crew-07Pending

02Scopetravel.bookingPending

03Cost$428.20 / $500Pending

04Data24h retentionPending

05ApprovalHuman requiredPending

DecisionREADY FOR PREFLIGHTAWAITING RESULT

How it works

Declare

Agent and service publish identity, intent, and hard boundaries.

Negotiate

Both sides resolve cost, scope, data rules, and approvals.

Act

An auditable decision and x402 settlement proof travel with the action.

Machine entry points - live now

One check, in the protocol your agent already speaks.

IntentFence publishes standard discovery files and callable endpoints so an agent runtime can find and enforce the service without reading this page.

A2A Agent Card IntentFence manifest x402 service manifest OpenAPI 3.1 llms.txt x402 payment metadata Receipt signing keys

01 / REST

POST /api/preflight

Free JSON policy preview for any runtime, workflow, or backend.

02 / x402

POST /api/preflight/verified

Pay 0.05 USDC per successful action with settlement proof and an ES256-signed receipt.

03 / MCP + A2A

Machine-native discovery

REST, MCP, A2A, OpenAPI, llms.txt, and Bazaar entry points.

60-second quickstartFREE REST

curl -X POST https://agentpass-protocol.rmalka06.chatgpt.site/api/preflight \
  -H "Content-Type: application/json" \
  -d '{
    "subject":"did:web:my-agent",
    "action":{"type":"purchase","resource":"order-42"},
    "constraints":{"cost_ceiling":100,"quoted_cost":79,"data_retention_hours":24}
  }'

Autonomous payment flowLIVE x402

POST /api/preflight/verified
-> 402 + PAYMENT-REQUIRED
-> agent signs 0.05 USDC on Base
-> retry + PAYMENT-SIGNATURE
-> 200 + PAYMENT-RESPONSE

Open protocol - autonomous USDC payments

Start per action. Upgrade when the policy becomes critical.

Agents can pay 0.05 USDC per signed decision. Teams that need an enforced production workflow can launch with a fixed-price integration pilot.

Open Preview

$0

free, unsigned preflight

Open spec and discovery files
REST, MCP and A2A
Deterministic policy checks
No wallet required

Open the API spec

Live pay-per-use

Verified x402

0.05 USDC

per action - live on Base

Agent pays directly
No account or API key
On-chain settlement proof
ES256-signed policy receipt
Machine-readable discovery
Payment audit record

Inspect the live payment flow

Launch Pilot

$750 + $149/mo

one guarded production workflow

Hands-on integration
Custom spend and approval policy
Guarded SDK wrapper
Receipt verification support
Founding-customer priority

Production

$499/mo

for teams operating agent fleets

Volume pricing
Team policy sets
Custom retention and SLA
Private deployment option

The 0.05 USDC endpoint is live. Pilot and Production prices are launch offers and include implementation support for an enforced workflow.

Founding customer program

Bring one agent action that can spend, send, deploy, or delete.

We will put IntentFence directly in the execution path, define the policy, and ship a guarded pilot around that workflow.

Why the world will need it

The web has rules for reading. Agents need rules for acting.

Autonomous software is crossing the line from answering questions to spending money, moving data, and committing people. IntentFence turns every action into a bounded contract before anything happens.

For agents

Know the rules before calling the tool.

Fewer failed actions, surprise charges, and unsafe assumptions.

For businesses

Publish boundaries once, in machine language.

Accept autonomous customers without surrendering control.

For people

Approve a contract, not a stream of clicks.

Every action stays explainable, limited, and auditable.

Open protocol / v0.5

A tiny manifest with a very big job.

Publish an IntentFence policy endpoint and wrap the downstream tool call with the SDK guard. The action runs only after an allowed decision.

DISCOVERGET /.well-known/intentfence.json

NEGOTIATEPOST /api/preflight

PAY + VERIFYPOST /api/preflight/verified

VERIFY RECEIPTPOST /api/receipts/verify

A2AGET /.well-known/agent-card.json

intentfence.jsonJSON

{
  "intentfence": "0.5",
  "subject": "did:web:crew-07",
  "action": {
    "type": "travel.booking",
    "resource": "TLV-LHR"
  },
  "constraints": {
    "currency": "USD",
    "cost_ceiling": 500,
    "data_retention_hours": 24,
    "human_approval": "required"
  },
  "decision": {
    "status": "safe_to_proceed",
    "quote": 428.20
  }
}

THE ACTION LAYER IS ARRIVING

Give every agent a safe way to say: “I’m allowed to do this.”

Choose a plan