Skip to main content
A request log records one model call. Use it to inspect exactly what your application sent and received, diagnose a failure, review latency and token usage, or connect production behavior to a prompt template. Each request log can include:
  • Model input and output
  • Provider and model
  • Start time, end time, and latency
  • Input and output tokens
  • Estimated cost
  • Status and error details
  • Prompt template association
  • Tags, metadata, scores, and a PromptLayer request ID

​
View Request Logs

Open Request Logs in the PromptLayer sidebar. The Requests table shows the model, prompt, status, timing, token usage, cost, and other fields for each call. Select a row to open its full details.
PromptLayer request logs table
To review logs for one prompt, open that prompt and select Analytics & Logs.
Request logs filtered to one prompt

​
Find a Request

Use the date range, free-text query, and structured filters above the Requests table. You can filter by fields such as prompt, model, provider, status, tags, metadata, content, latency, tokens, and cost. See Search for the dashboard workflow, supported filter categories, and older-data behavior.

​
Request Logs and Traces

A request log describes one model call. A trace describes the complete operation around that call and can contain multiple request logs. When PromptLayer receives a supported GenAI span, it creates a request log for the model call and links it to the trace. If PromptLayer.run() already created the request log, PromptLayer links the span to that log instead of creating a duplicate.

​
Configure Request Logs

Use tags, metadata, and scores to make requests easier to find and compare. Every log also has a PromptLayer request ID that you can store or use with tracking methods. Use Tracking Templates to associate provider calls with PromptLayer prompts. For manual provider calls or unsupported models, use Custom Logging. If a provider returns JSON, see Structured Outputs.

​
Reuse Production Data

Select request logs from the table and add them to a PromptLayer Table. This lets you turn representative production examples or failures into evaluation and backtesting data.