Raidu is the thing your auditors will use to verify your AI. So we hold ourselves to the same bar. This page is the source of truth. It updates when we do.
Every execution produces a signed document. Raidu publishes its verification public key. Anyone with the record and the key can confirm, offline, that it has not been altered. That is the bar. Not logs-in-a-bucket. Cryptographic proof.
{
"schema": "raidu.record/v2",
"record_id": "rec_01JBVX7P9A8Z8PTQJG4K9NDJ4W",
"tenant": "acme-corp",
"agent": "support-triage-v3",
"model": "anthropic/claude-sonnet-4.5",
"started_at": "2026-04-21T14:02:11.041Z",
"checkpoints": [
{ "ck":"01", "name":"user_input", "latency_ms":2.1,
"entities": ["EMAIL","US_SSN","CREDIT_CARD"] },
{ "ck":"02", "name":"before_llm", "latency_ms":3.4,
"masked": 3, "policy":"enterprise.v17" },
{ "ck":"03", "name":"before_tool", "latency_ms":2.6,
"tool":"gmail.send", "outbound_pii":"scoped" },
{ "ck":"04", "name":"after_tool", "latency_ms":1.8,
"new_entities": 0 },
{ "ck":"05", "name":"agent_response","latency_ms":2.2,
"final_scan":"clean" }
],
"policies": ["hipaa.baa", "eu-ai-act.art10+13", "no-card-to-llm"],
"decision": "allow",
"explanation":"PII masked before model. Tool call scoped.
Response clean. No policy violations.",
"prev_hash": "b7f3…c421",
"hash": "e9a2…d3fa",
"rfc3161_ts": "2026-04-21T14:02:11.051Z",
"signature": "MIIFxjCCA6… (RSA-4096, 684 bytes) …Qw=="
}Raidu's design principle is that no third party, including our own vendors, ever sees unmasked AI payloads. The infrastructure below runs the service; it does not see the substance of what crosses the runtime.