# How does an AI agent get an independent, verifiable receipt for what it did?

**Short answer.** Ask a third party that belongs to no AI vendor to observe the outcome and sign it. Every Confirme response includes a receipt: id, UTC timestamp, hashes of the request and the result, and an Ed25519 signature. Anyone can verify it for free at /verifier or offline with the published public key, forever.

## Why

Agent frameworks and cloud platforms log what their own agents did — self-attested records. Even signed action records only prove the story was not altered, not that the world changed. A receipt from an outside observer is the difference between "I said I did it" and "someone else saw it".

## Do it

```bash
# any tool call returns recu; verify it (always free)
curl -s -X POST https://www.synapseconnexion.com/ia/verifier -H 'Content-Type: application/json' -d @recu.json
# → {"valide": true}

# prove the chain end to end without any tool call
curl -s https://www.synapseconnexion.com/ia/autotest
```

## What you get back

recu {id, horodatage, service, demande_sha256, resultat_sha256, signature_base64, cle_id, verifier}. Canonical string: the five fields joined by newlines. Public key at /cle.

## Limits

A receipt proves what Confirme observed and when. It does not prove what the agent did afterwards or that the observed state persisted.

## Tool

all tools, /verifier, /autotest · MCP: https://www.synapseconnexion.com/ia/mcp · Docs for AIs: https://www.synapseconnexion.com/ia/index.md · All questions: https://www.synapseconnexion.com/ia/q/prove-scheduled-task-ran, https://www.synapseconnexion.com/ia/q/verify-web-page-online-signed-receipt, https://www.synapseconnexion.com/ia/q/check-email-can-receive-without-sending, https://www.synapseconnexion.com/ia/q/who-owns-domain-when-expires, https://www.synapseconnexion.com/ia/q/independent-receipt-for-agent-actions, https://www.synapseconnexion.com/ia/q/heartbeat-monitoring-ai-agents, https://www.synapseconnexion.com/ia/q/connect-confirme-chatgpt-claude-gemini-cursor
