Usage logs

Usage logs

Use Usage logs to verify the model that served a request, inspect recorded tokens and costs, and investigate client configuration. Payment orders are tracked separately from model requests.

Find a request

  1. Send one short prompt from the application you are configuring.
  2. Open Usage logs and select a time range that includes the request.
  3. Narrow the results using the available model, key, group, or type filters.
  4. Open the matching record and compare the timestamp, model, tokens, and charge.

Understand the recorded fields

FieldHow to use it
Time, model, and keyMatch the log entry to the test request and application.
Input and output tokensUnderstand prompt size, generated output, and model usage.
Cache read and write tokensInspect provider-reported cache usage when available. Missing data is different from a confirmed zero.
Group and billing informationCheck the route and rate settings associated with the request.
Latency and durationCompare responsiveness while accounting for prompt size, output length, and streaming.
Recorded costReconcile the request with the applicable catalog rate and billing source.

Streaming and cache usage

Full usage may arrive near the end of a stream. Do not calculate the final charge from the first chunk. A partial response, interrupted connection, or missing usage field does not prove that the upstream performed no work.

Cache fields depend on the model and upstream response. Do not add cached tokens to input tokens again unless the selected protocol defines them as separate quantities; some APIs report cached input as a subset.

When a request is missing

Refresh the logs and widen the time range, then verify that the application uses a LivRouter key and the correct endpoint. Requests rejected before a model call may not produce a normal consumption record. Check the client's error body and the troubleshooting guide.

For payment receipts, account balance, and subscriptions, open Billing & credits.