Skip to main content
Planck captures model activity so you can investigate failures, measure cost and latency, and understand how an application behaves in production.

Properties

Attach environment, feature, customer, or workflow metadata to requests.

Sessions

Group related requests into conversations or multi-step agent runs.

User Metrics

Attribute usage, cost, and behavior to application users.

Datasets & Reports

Build reusable request sets and analyze them over time.

Scores & Feedback

Record automated evaluations and end-user feedback.

Webhooks

Send matching Planck events to another service.

​
A practical setup

  1. Add a stable user ID and session ID to each request.
  2. Add a small set of consistent properties such as environment, feature, and workflow.
  3. Record feedback or scores against the request that produced the output.
  4. Use datasets and reports for repeatable analysis.
Keep property names stable and values low-cardinality when possible. Consistent metadata makes filtering and comparisons much easier.
For programmatic access, see the Observability API.