Skip to content
OnCoreDocs

Calls

Read call activity

Read call activity: filtered lists, statuses, per-assistant summaries, bulk export, full single-call payloads, and per-call session insights.

The calls API is your read-only window into everything your assistants have handled on the phone: transcripts, summaries, analysis, cost, and recording links.

Read-only

There is no API to place calls. Calls are created by the platform when someone dials one of your phone numbers; these endpoints only read the results.

MethodPathPurpose
GET/callsList calls with filters and pagination
GET/calls/statusesDistinct call statuses seen in your tenant
GET/calls/summary-by-assistantPer-assistant call summaries
GET/calls/exportBulk export (up to 10,000 rows, lean fields)
GET/calls/{id}One call with the full payload
GET/calls/{id}/insightsSession insights for one call: models, per-turn latency, tool timeline
GET/calls/insightsSession insights for up to 25 calls in one request

List calls#

GET /calls

Query paramTypeRequiredDescription
limitnumberNoPage size. Default 50, maximum 500.
offsetnumberNoItems to skip. Default 0.
assistantIdUUIDNoOnly calls handled by this assistant
statusstringNoFilter by status (values from /calls/statuses)
dateFrom / dateToISO 8601 datetime (with offset)NoTime window, e.g. 2026-07-01T00:00:00+00:00
durationMin / durationMaxnumberNoCall duration bounds
sortBystartedAt | createdAtNoSort key (default startedAt)
searchstringNoFree-text search
bash
curl "https://core-api.heysadie.ai/calls?assistantId=ASSISTANT_ID&dateFrom=2026-07-01T00:00:00%2B00:00&limit=100" \
  -H "Authorization: ApiKey YOUR_API_KEY"

Discover statuses#

GET /calls/statuses returns the distinct status strings that actually occur in your tenant's call history. Use it to build filter dropdowns rather than hard-coding values.

Terminal
curl https://core-api.heysadie.ai/calls/statuses \
  -H "Authorization: ApiKey YOUR_API_KEY"

Per-assistant summaries#

GET /calls/summary-by-assistant returns one row per assistant:

FieldDescription
assistantId / assistantNameWhich assistant
totalCallsTotal call count
avgDurationAverage duration (nullable)
successRateSuccess rate
Terminal
curl https://core-api.heysadie.ai/calls/summary-by-assistant \
  -H "Authorization: ApiKey YOUR_API_KEY"

Export calls in bulk#

GET /calls/export returns up to 10,000 calls as JSON in one response. It accepts the same filters as the list endpoint (assistantId, status, dateFrom, dateTo, durationMin, durationMax, search) but no pagination params.

Lean rows only

Export rows exclude the heavy fields: no transcript, messages, analysis, summary, or recording URL. You get identifiers, timing, status, cost, and duration. Fetch GET /calls/{id} for the full payload of any call you need in depth. Use date-range filters to stay under the 10,000-row cap on busy tenants.

Terminal
curl "https://core-api.heysadie.ai/calls/export?dateFrom=2026-07-01T00:00:00%2B00:00&dateTo=2026-08-01T00:00:00%2B00:00" \
  -H "Authorization: ApiKey YOUR_API_KEY"

Get one call#

GET /calls/{id} returns the full call record:

FieldDescription
id, cidCall identifiers
type, status, endedReasonWhat kind of call, how it ended
startedAt, endedAt, callDurationTiming
transcript, messagesFull transcript and message log
summary, analysisAI-generated summary and structured analysis
recordingUrlRecording link (when recording is enabled)
cost, costBreakdownBilling detail
customerNumberThe caller's number
assistantId, assistantNameWhich assistant handled it
supervisorAnsweredWhether a human supervisor answered (transfers)
createdAt, updatedAtRecord timestamps
Terminal
curl https://core-api.heysadie.ai/calls/CALL_ID \
  -H "Authorization: ApiKey YOUR_API_KEY"

Read session insights#

GET /calls/{id}/insights returns how one call actually ran. It joins the call record with the performance metrics the voice pipeline recorded during the call. id is the same call id GET /calls/{id} takes.

FieldDescription
idThe call id, the same value the call record carries
startedAt, endedAt, callDurationTiming
hasRecordingWhether a recording exists. The URL itself is on GET /calls/{id}
pipelineOne entry each for eou, llm and tts: the model that ran that stage and its avgMs over the call (end-of-utterance delay, LLM time to first token, TTS time to first byte)
summaryTurn counts, tool call and tool failure counts, interruptions, the average latencies in milliseconds, and agentProcessingMs, the average latency multiplied by the agent turn count
performanceOne entry per assistant turn: e2eLatencyMs, llmTtftMs, ttsTtfbMs, transcriptionDelayMs, endOfUtteranceDelayMs, interrupted, token counts and cache status
timelineEvents in call order, each with offsetMs from the start of the call: user_speech, agent_response, tool_call and tool_result

Every latency is in milliseconds. A value the pipeline did not measure is null, never 0, and the averages in summary are null when no turn carried the value. agent_response.latencyMs is that turn's end-to-end latency. tool_result.result is the raw body your server returned, as a string, and tool_result.error is set when the call failed.

Metrics land a short while after the call ends. Until they do, this endpoint answers 404 with the message Call insights not found, even though GET /calls/{id} already returns the call. Retry after a minute rather than treating that first 404 as final. A 404 with Call not found is different: the id is not a call in your tenant.

bash
curl https://core-api.heysadie.ai/calls/CALL_ID/insights \
  -H "Authorization: ApiKey YOUR_API_KEY"

Read insights for several calls#

GET /calls/insights?ids=... returns the same insights for up to 25 calls in one request.

Query paramTypeRequiredDescription
idscomma-separated UUIDsYes1 to 25 call ids, no duplicates. Spaces around the commas are ignored

The response has two arrays. insights holds one entry per call that produced insights, in the order you sent the ids. missing names every id that produced nothing, with a reason: call_not_found when the id is not a call in your tenant, metrics_not_found when the call exists but has no metrics yet. One unknown id never fails the request, and a request where every id is missing is still a 200.

bash
curl "https://core-api.heysadie.ai/calls/insights?ids=0195f1e2-0000-0000-0000-000000000000,0195f1e2-0000-0000-0000-000000000001" \
  -H "Authorization: ApiKey YOUR_API_KEY"

Both insights endpoints answer 503 with code PERFORMANCE_SERVICE_UNREACHABLE if the metrics store cannot be reached. Retry with backoff; the call record itself stays available on GET /calls/{id}.

Next steps#

PreviousSearch, buy and configureNextRecording & transcripts