Traces

Where a call's time and cost actually went.

A trace is every step behind one piece of work (the model calls, the tool calls, the retries) with its timings, tokens and cost. Each names the call, agent, campaign or eval run it belongs to, so it joins to the rest of this API rather than sitting in its own world.

Which vendor and which model served each step are not reported. Those change (that is the point of running several) and a dashboard grouped by a model name would break the day one moved. What is reported is the work: how long it took, what it consumed, and whether it failed. A model call's span is named for the operation (chat, stt), and an upstream error whose text names a vendor is reported as a neutral message.

Span payloads (prompts, model output, tool arguments) are not served here. They are the most sensitive thing we hold, since a prompt carries whatever the call was about, and they stay behind the signed-in surface where the reader is identifiable.

List traces#

Traces in the organization, newest first. A default time window applies when none is given. An unbounded scan of everything ever produced is not a query to run by accident.

GET/api/v1/traces

Query parameters

fromstring: ISO date lower bound.
tostring: ISO date upper bound.
callIdstring: Traces for one call.
campaignIdstring: Traces for one campaign.
agentIdstring: Traces for one agent.
evalRunIdstring: Traces for one eval run.
sourcestring: The kind of work: voice, analysis, data_collection, objective, campaign_analysis, call_center_analysis, help_desk, eval, report, agent_compose or other. Comma-separate for several. Any other value is refused.
statusstring: ok or error.
hasErrorstring: `1` for failures only.
minDurationMsnumber: Slow work only.
qstring: Free text over the trace name.
sortstring: startedAt, durationMs, costUsd, totalTokens or errorCount. Prefix with - for descending.; Default-startedAt
pagenumber: Page number.; Default1
pageSizenumber: Up to 200.; Default50

Responses

A page of traces.

Request

curl -X GET "https://voice.evryhealth.com/api/v1/traces" \
  -H "Authorization: Bearer YOUR_API_KEY"

Response200

{
  "data": {
    "traces": [
      {
        "id": "7a40…",
        "name": "call.analysis",
        "source": "analysis",
        "status": "ok",
        "durationMs": 4182,
        "links": {
          "callId": "5f1c…",
          "campaignId": "a1…",
          "agentId": "44b1…",
          "callCenterId": null,
          "evalRunId": null
        },
        "counts": {
          "spans": 14,
          "llmCalls": 9,
          "toolCalls": 2,
          "errors": 0
        },
        "usage": {
          "inputTokens": 18422,
          "outputTokens": 2201,
          "totalTokens": 20623,
          "costUsd": 0.0412
        },
        "tags": []
      }
    ],
    "total": 318
  },
  "meta": {
    "requestId": "req_…",
    "pagination": {
      "nextCursor": "2",
      "limit": 50
    }
  }
}

Get a trace#

One trace and every span under it, in order: the waterfall for one piece of work.

GET/api/v1/traces/{traceId}

Path parameters

traceIdstringrequired: Trace id.

Responses

The trace and its spans.

Request

curl -X GET "https://voice.evryhealth.com/api/v1/traces/{traceId}" \
  -H "Authorization: Bearer YOUR_API_KEY"

Response200

{
  "data": {
    "trace": {
      "id": "7a40…",
      "name": "call.analysis",
      "status": "ok",
      "durationMs": 4182
    },
    "spans": [
      {
        "id": "b112…",
        "parentSpanId": null,
        "depth": 0,
        "kind": "llm",
        "name": "sentiment",
        "status": "ok",
        "durationMs": 812,
        "timeToFirstTokenMs": 240,
        "toolName": null,
        "usage": {
          "inputTokens": 2201,
          "outputTokens": 180,
          "totalTokens": 2381,
          "costUsd": 0.0041
        }
      }
    ]
  },
  "meta": {
    "requestId": "req_…"
  }
}