Calls

Setup steps and authentication are in the Overview. This page covers production-call observability.

A call is a production conversation — captured from your live voice agent. Once a call is in Cekura, you can run metrics on it, build test sets from it, and analyze trends across thousands of calls. Webhooks from Vapi / Retell / ElevenLabs / LiveKit / Pipecat ingest automatically; you can also send calls programmatically.

List calls

cekura calls list --project-id 742
cekura calls list --agent-id 123 --format json | jq '.[] | .id, .duration'

calls = client.calls.list(project_id=742, page_size=50)
for c in calls.get("results", calls):
    print(c["id"], c.get("duration"), c.get("call_ended_reason"))

Common filters: project_id, agent_id, from_date, to_date, success, topic. See the API Reference for the full list.

Inspect a call

cekura calls get 7788

call = client.calls.get(call_id=7788)
print(call["transcript"])
print(call["metric_evaluations"])

Send a call for evaluation

If your provider isn’t on the auto-ingestion list — or you want to ship a call from your own backend — send it explicitly.

cekura calls send --from-file call.json

Where call.json contains the agent ID, transcript, duration, and metadata.

client.calls.send(
    agent=123,
    transcript=[
        {"role": "agent", "text": "How can I help?"},
        {"role": "user",  "text": "I'd like to reschedule my appointment."},
    ],
    duration=124,
    metadata={"source": "internal-bot"},
)

Provider-specific webhook helpers are also available:

client.calls.observe_vapi(payload=...)
client.calls.observe_retell(payload=...)
client.calls.observe_elevenlabs(payload=...)
client.calls.observe_livekit(payload=...)
client.calls.observe_pipecat(payload=...)

In production these are usually wired as webhooks pointing at your backend or directly at api.cekura.ai/observe — the SDK helpers are useful for replays and testing.

Run metrics on a call

Score an already-ingested call against one or more metrics.

cekura calls evaluate 7788 --metric-ids 55,56

# Re-run a previous evaluation (e.g. after editing a metric prompt)
cekura calls rerun-evaluation 7788 --metric-ids 55

client.calls.evaluate_metrics(call_log_id=7788, metric_ids=[55, 56])

# Re-run a previous evaluation
client.calls.rerun_evaluation(call_log_id=7788, metric_ids=[55])

Generate scenarios from real calls

Found an interesting production call? Turn it into regression scenarios.

cekura calls create-scenarios 7788 --count 3
cekura calls create-scenarios-progress --progress-id <id>

job = client.calls.create_scenarios(call_log_id=7788, count=3)
client.calls.create_scenarios_progress(progress_id=job["progress_id"])

Promote a call into a test set

cekura test-sets create-from-call-log \
  --call-log-id 7788 \
  --name "support-edge-cases"

client.test_sets.create_from_call_log(
    call_log_id=7788,
    name="support-edge-cases",
)

Vote on a metric result

Capture thumbs up/down feedback on a specific metric evaluation for a call, optionally attach the expected value and free-text feedback. The call is marked as reviewed; the metric evaluation is updated. Feeds the labs / metric-review workflow.

# Thumbs up — agreement with the metric's verdict
cekura calls mark-metric-vote 7788 \
  --metric-id 55 \
  --thumbs-up \
  --expected-value 5 \
  --feedback "Agent greeted correctly"

# Thumbs down — disagreement, supply what you'd have expected
cekura calls mark-metric-vote 7788 \
  --metric-id 55 \
  --thumbs-down \
  --expected-value 2 \
  --feedback "Greeting was too curt"

--expected-value is parsed as JSON when possible (5, true, "foo"), so numeric / boolean metrics get the right type.

client.calls.mark_metric_vote(
    call_log_id=7788,
    metric_id=55,
    thumbs_up=True,
    expected_value=5,
    feedback="Agent greeted correctly",
)

Flag a critical-scenario verdict as wrong

If a critical-scenario evaluation looks incorrect, flag it (and later unflag if it was a misclick or has since been corrected).

client.calls.mark_critical_scenario_wrong(call_log_id=7788, scenario_id=42)
client.calls.unmark_critical_scenario_wrong(call_log_id=7788, scenario_id=42)

Metrics

Define what you score production calls on.

Runs & Results

Same scoring engine — for simulated, not production, traffic.

Dashboards

Visualize call quality and metric trends over time.

API Reference

Full field reference for call logs and observation payloads.

CLI & SDK

List calls

Inspect a call

Send a call for evaluation

Run metrics on a call

Generate scenarios from real calls

Promote a call into a test set

Vote on a metric result

Flag a critical-scenario verdict as wrong

See also

Metrics

Runs & Results

Dashboards

API Reference

​List calls

​Inspect a call

​Send a call for evaluation

​Run metrics on a call

​Generate scenarios from real calls

​Promote a call into a test set

​Vote on a metric result

​Flag a critical-scenario verdict as wrong

​See also

Metrics

Runs & Results

Dashboards

API Reference

List calls

Inspect a call

Send a call for evaluation

Run metrics on a call

Generate scenarios from real calls

Promote a call into a test set

Vote on a metric result

Flag a critical-scenario verdict as wrong

See also