Skip to main content

Logs and usage

Logs shows each request your keys send. Usage shows totals for a period. Sign into the console with the account that owns the key.

Find a request in Logs​

Requests appear in Logs within a minute, newest first, with times in UTC.

  1. Select Logs.
  2. Set Model (or All models), Period (up to Last 30 days) and Status (All statuses, Succeeded or Failed).
  3. If your request isn't shown, select Load older requests.

Each row shows the requested model, endpoint, status, tokens and latency. The token count covers input and output, plus any cached, cache-write and reasoning tokens.

Select View request or View response to read the content. The status Streamed; not stored means Divyam streamed the answer without saving it. Not recorded means Divyam didn't save the content.

Review usage​

Usage catches up within a few minutes.

  1. Select Usage, then choose a period (Last hour to Last 90 days, or Custom range) and a model.
  2. Read the cards: Requests, Total tokens, Success rate and Spend. Divyam doesn't charge failed requests.
  3. Toggle the chart between Requests, Tokens and Spend, and between Combined and By model.

The By requested model table shows requests, success rate, tokens and spend for each model.

Cached tokens​

A model can reuse a prompt prefix it has seen before. Tokens read from that cache cost less than fresh input. The Tokens chart splits your tokens into bands that add up to Total tokens:

  • Uncached input: prompt tokens processed fresh.
  • Cached input: prompt tokens read from the cache.
  • Cache write: prompt tokens stored in the cache for reuse.
  • Output excl. reasoning: the answer.
  • Reasoning: tokens the model used to think before answering.

Input tokens counts all three input bands, and Output tokens includes reasoning. Cached input and cache writes have their own rates, uncached input uses the Input rate, and reasoning uses the Output rate. See Models and pricing.

Requested against routed​

Your appCoding harnessDivyamrouterModel AModel BModel C
Your app or coding harness asks for one catalog model, and Divyam picks the backing model for each request. Usage shows the split.

Your application requests a catalog model, and Divyam may answer with a different backing model. Requested against routed, on Usage, shows how Divyam split each model's requests across backing models, failed requests included. Select Show the flows for request counts and shares.

Logs and the response's model field show only the requested model. If Divyam recorded no routed model for the period, the view says so and can't show the split.