How to prove what your AI agent did

Identity, mandate, log book, passport, Evidence Pack. Each step is something the other side can check.

1. 1. Give the agent an identity

   Sign in and register the agent at apir.ai/agents/connect. It can run anywhere: your own code, any model, any framework. It gets an API key and signed papers, an Ed25519 credential under did:web:apir.ai. The first agent is free; each further agent is AUD 49, once, for life.

2. 2. Write its mandate

   Say what the agent may do, what it may never do, and its limits. You write it; it goes on the record with the agent. From then on, "did it stay inside its authority?" is a question the record answers.

3. 3. Record every action in the Black Box

   Each time the agent acts, decides or calls a tool, send one event. Hash the prompt and the output on your side and send only the SHA-256 hashes. Up to 50 events go in one request.

   POST https://apir.ai/api/v1/events
   Authorization: Bearer <the agent's API key>
   Content-Type: application/json
   
   {
     "agent_id": "<agent id>",
     "events": [
       { "type": "decision", "summary": "Approved a refund within its mandate",
         "input_hash": "<sha256 of the prompt>", "output_hash": "<sha256 of the answer>" }
     ]
   }

4. 4. Let it earn its Trust Passport

   After five conduct events (actions, decisions, tool calls) the Trust Genome grades the agent from its record, and it is re-graded nightly. At 70 or more it is issued an Ed25519-signed Trust Passport. An agent with no evidence reads Unrated, never a made-up score. A public passport has a page anyone can open, listed in the registry.

5. 5. Hand over an Evidence Pack

   When a deal, a security review or a claim needs it, seal one agent's record into an Evidence Pack: the hash-chained history, sealed and Ed25519-signed, behind a public verify link. The other side opens it with no account and checks it themselves. AUD 499, one-time, no subscription. See pricing.

They never have to take your word, or APIR's.

• Paste the agent's name or passport ID at apir.ai/check-agent: the signature is re-checked against the published key, the mandate read back and the log book recomputed. Free, no account.
• Check the signature offline against apir.ai/.well-known/jwks.json with any Ed25519 library.
• Open the Evidence Pack's verify link: no account, no login.
• Ask an AI assistant: APIR's MCP server (ai.apir/verify, https://apir.ai/api/mcp) runs the same check.

It proves what the agent reported and that nobody changed it afterwards.

It does not prove the report was complete: an agent that never sends an event leaves no entry. It does not prove the agent was right, or safe. The mandate is yours, so the record shows whether the agent kept to it, not whether it was a good mandate. Whether a regulator, a court or an insurer accepts the record is their call; APIR gives no legal advice.

We don't promise results. We prove whatever happens.

Checking is free, for anyone, forever.

• The check at apir.ai/check-agent: free, no account, never paywalled.
• The first papered agent (the First Employee): free.
• Each further agent's papers: AUD 49, once, for life.
• Wage: AUD 99 per agent, per month. Crew: AUD 499 per month, up to ten agents.
• Evidence Pack: AUD 499, one-time.

Prices are read from the rate card at apir.ai/pricing when this page is built.

How do I prove what my AI agent did?

Keep a record someone else can check: signed papers for the agent, a written mandate, an insert-only hash-chained log book of every action, and an outside timestamp on that log. With APIR: register the agent at apir.ai/agents/connect, write its mandate, send each action to POST https://apir.ai/api/v1/events, let it earn a Trust Passport, and hand over an Evidence Pack with a public verify link.

Can I prove what my agent did without sharing its prompts?

Yes. Hash each prompt and output on your side and send only the SHA-256 hashes. Anyone holding the original text can show it matches the hash on the record; nobody can read the text from the hash.

Does it work with agents built on OpenAI, Anthropic, LangChain or my own code?

Yes. Any agent that can make an HTTPS request can send events to POST https://apir.ai/api/v1/events with its API key. The model and the framework do not matter.

What does a buyer or an insurer get?

A link they open with no account: the agent's signed papers, its mandate and its log book, recomputed as they open it, or an Evidence Pack sealing one agent's record for a deal, review or claim. APIR is not an insurer or a broker and certifies no one; it keeps the record.

How much does it cost?

Checking an agent is free. The first papered agent is free. Each further agent's papers are AUD 49, once. An Evidence Pack is AUD 499, one-time. Wage is AUD 99 per agent, per month, and Crew is AUD 499 per month for up to ten agents. Prices are on the rate card at apir.ai/pricing.

Is an APIR log book legal proof?

It is evidence that anyone can verify independently: signed, hash-chained and timestamped by an outside authority. Whether a court, a regulator or an insurer accepts it is for them to decide. APIR gives no legal advice.