Prompt Trace + Diff
Every intent call produces a full execution trace automatically — prompt, retrieved context, guardrail checks, and output.
Every intent call produces a full execution trace — automatically, with no instrumentation required. See the exact prompt sent to the model, every retrieved chunk with its similarity score, and every guardrail decision, all in one queryable place.
What's captured
Full Prompt Visibility
The exact prompt sent to the model — after retrieval, after templating — not what you think you sent.
Retrieved Context
Every chunk retrieved, with source, similarity score, and the exact text passed to the model.
Guardrail Decisions
Which policy checks ran and whether they passed, flagged, or blocked — for every call.
Cost & Latency Per Stage
Each stage — prompt, retrieval, guardrails, output — reports its own contribution, not just a total.
Reading a trace
Every /v1/run response includes a metadata block with the trace summary; the full trace (retrieved chunks, guardrail evaluations, per-stage timing) is queryable via the dashboard or the traces API:
GET /dashboard/traces/:messageId
Authorization: Bearer <jwt>{
"success": true,
"data": {
"trace": {
"messageId": "msg_xyz",
"stages": [
{ "stage": "prompt", "latencyMs": 4 },
{ "stage": "retrieval", "latencyMs": 38, "chunks": [{ "source": "policy.md", "score": 0.94 }] },
{ "stage": "guardrails", "latencyMs": 11, "checks": [{ "name": "pii_redaction", "result": "passed" }] },
{ "stage": "output", "latencyMs": 2 }
],
"totalLatencyMs": 340,
"costUsd": "0.0021",
"model": "claude-sonnet-5"
}
}
}Traces are queryable indefinitely after the fact — reproduce a bad output weeks later without needing to have been watching in real time.