Agent accountability

What did your AI agent
actually do?

Nobody can answer that today. Forge Box issues an ed25519-signed receipt for every agent execution: what ran, what it cost, what policy allowed. Verifiable offline, forever.

Get audit-ready free See a live receipt

Security / compliance leads

Receipts from our own production runs:

{
  "id": "rcpt_797a7ff6d425",
  "action": "workflow_execution",
  "risk_score": 10,
  "cost_usd": 0.001,
  "timestamp": "2026-09-17T20:49:50.138Z",
  "algorithm": "ed25519",
  "key_id": "key_2b47722ac3eed01a",
  "signature": "RKSobO4a1Y/sBv27ivpb5GNM..."
}

Cryptographic proof

Canonical JSON + ed25519 signature. Byte-exact — one changed field and verification fails. Not "trust our logs."

Policy before the action

Spend limits, allowed domains, risk scoring 0-100. Guardrails evaluate before anything executes — denials are recorded too.

Auditor-friendly

Public signing key at a stable endpoint. Your auditor verifies receipts years from now without ever calling your API vendor.

Live

Run a sandbox action. Get the real receipt.

This is the production sandbox — same infrastructure, same signing key, no simulation.

Press "Run it" — output, risk score, and receipt signature appear here.