API for AI agent accountability

Your agent did it.
Now prove it.

AI agents negotiate prices, cancel subscriptions, book travel and place orders on your behalf. AgentProof turns each of those actions into a timestamped, tamper-evident receipt — one anyone can verify.

Request early access Read the API

The gap between what happened and what you can prove

Agents are getting good at acting. They are not getting good at testifying. When the other side says "we have no record of that", an agent's word is worth nothing — on both sides of the table.

CASE 01

The discount that vanished

Your agent negotiated 30% off. The next invoice arrives unchanged, and nobody can find the conversation.

CASE 02

The subscription that came back

The agent cancelled it. You were charged again the following month — and the confirmation screen is long gone.

CASE 03

The price that never was

The agent saw $89. Later the page shows $129 and the "deal" is called a misread.

How it works

Three calls. No agent framework to adopt, no data to hand over.

STEP 1

Report the action

Your agent posts the outcome — what it did, with what values, at what time. A single HTTP call.

STEP 2

Get a receipt

We return a receipt ID, a content hash and a verification URL. The hash is anchored so it cannot be rewritten later.

STEP 3

Show it, verify it

Paste the receipt anywhere. Anyone holding the original record can recompute the hash and confirm it matches — or see that it doesn't.

What makes it evidence rather than a log: a log lives on someone's server and can be edited by whoever runs that server. A receipt is a hash anchored outside both parties' control. Neither side can quietly change the past.

Where it earns its keep

Negotiation records

Commitments made by a human agent or a support bot, frozen at the moment they were made.

Cancellation proofs

Proof that a cancellation was submitted and acknowledged — useful when the charge reappears.

Price and quote snapshots

What the page said, when it said it. Makes "dynamic pricing" auditable.

For developers

A single REST endpoint. Works with any agent stack — no SDK required, and no lock-in.

POST https://api.proofofaction.dev/v1/receipts
Authorization: Bearer $AGENTPROOF_KEY

{
  "agent_id":   "support-bot-01",
  "action":     "subscription.cancelled",
  "occurred_at": "2026-09-23T14:02:11Z",
  "evidence":   { "confirmation_id": "CN-88213", "channel": "web" }
}

→ 200
{
  "receipt_id":  "rcpt_7Qm2Kd9",
  "content_hash": "9f2c...a1",
  "verify_url":   "https://proofofaction.dev/v/rcpt_7Qm2Kd9",
  "anchored_at":  "2026-09-23T14:02:13Z"
}

Also available as an MCP server if your agent speaks the Model Context Protocol, and as a Muse connector for agents running on Meta's Muse.

Built data-minimal on purpose

We designed this so that handing us an action does not mean handing over your data.

We store hashes, not content. The values you send are hashed with a per-account salt; the raw record stays on your side.
We never ask for credentials or access. No account delegation, no scraping, no standing connection to your systems.
You send, we don't fetch. Receipts exist only because your agent chose to report an action.
No sale, no brokering. Your records are not shared or monetized.
Deletion is real. Destroy the salt and the hash becomes unlinkable to any content, permanently.

Start with one call

Early access is open to developers building agents that act on a user's behalf. Tell us what your agent does and we'll get you a key.

Request early access

Your agent did it.Now prove it.