Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.cekura.ai/llms.txt

Use this file to discover all available pages before exploring further.

Setup steps and authentication are in the Overview. This page covers metric authoring and management.
A metric scores a run or production call. Cekura supports three flavors:
  • Predefined — platform-managed metrics like sentiment, interruption count, latency.
  • LLM judge — you write a prompt; an LLM scores the transcript.
  • Custom code — Python that runs against the call payload.

Browse predefined metrics

These are read-only and shared across the platform.
cekura predefined-metrics list

Create a metric

Prepare metric.json:
{
  "agents": [123],
  "name": "Booking confirmed",
  "type": "llm_judge",
  "eval_type": "boolean",
  "prompt": "Did the agent confirm a booking and read back the date and time?"
}
Apply:
cekura metrics create --from-file metric.json

List, update, delete

cekura metrics list --agent-id 123
cekura metrics get 55
cekura metrics update 55 --from-file patch.json
cekura metrics delete 55

Bulk operations

Manage many metrics in one call.
# Create several at once
cekura metrics bulk-create --from-file metrics.json

# Bulk-attach metrics across agents
cekura metrics bulk-manage-agents \
  --metric-ids 55,56 \
  --agents-to-add 123,124

# Toggle active state on many metrics
cekura metrics bulk-toggle-settings --metric-ids 55,56 --is-active true

Generate metrics with AI

Stuck on what to measure? Cekura can propose metrics from your agent’s scenarios.
cekura metrics generate --agent-id 123 --count 10
cekura metrics generate-progress --progress-id <id>

Critical metric scenarios & reviews

Two advanced workflows:
  • Critical metric scenarios — scenarios flagged because their metric output drifted. Useful for triage.
  • Metric reviews (Labs pipeline) — kick off feedback processing on metric judgments and watch progress.
# Critical metric scenarios
cekura critical-metric-scenarios list --metric-id 55
cekura critical-metric-scenarios update 42 --json '{"is_resolved": true}'

# Metric reviews
cekura metric-reviews process-feedbacks --metric-id 55
cekura metric-reviews progress --job-id <id>

See also

Runs & Results

Where metric scores show up after a run.

Calls

Run metrics against production calls, not just simulations.

Metric concepts

LLM judge, Python, rubric, sampling — when to use what.

API Reference

Full field reference for metric payloads.