Skip to main content
Guides

Reading your usage and charges

What the AICredits usage page shows for each request, how cached tokens appear, how long request content is kept, and why an old charge can differ from today's model price.

Use this page with an AI assistant

Opens a new chat with this docs URL and the correct AICredits base URLs.

Open Dashboard → Usage to see what each request used and what it cost.

The usage list

Choose a range (last 7, 30 or 90 days). Each row shows:

ColumnMeaning
TimeWhen the request was made (IST)
ModelThe model id. A "Cached" badge means the answer was served from a cache and cost nothing
TokensTotal tokens. Hover to see input and output separately. A line such as "1.2K cached" means part of the input was read from or written to the provider's prompt cache
Cost (INR)The amount taken from your wallet for that request
WalletPersonal, or Team if the request was charged to a team allocation

Details for one request

Click a row to open its details:

  • Overview: model, time, State, Finish reason, speed figures (time to first token, generation time) and routing, where available.
  • Tokens: Prompt, Completion, Cached read, Cached write, Reasoning and Total.
  • Cost: the final charge in rupees.
  • Content: the prompt and the response text, with a copy button.

Tokens in plain words

  • Prompt tokens are what you sent. Completion tokens are what the model wrote back.
  • Cached read tokens are input the provider had already seen and reused. Cached write tokens are input stored for reuse. See Prompt caching.
  • Reasoning tokens are the model's hidden thinking on reasoning models. See Reasoning tokens.

State and finish reason

State tells you whether the call completed or failed. A failed request that produced nothing has a zero charge. A request you cancelled partway is charged for what was generated. Details and the full list of cases are in Why a call failed, and whether you were charged.

How long things are kept

Request and response content is kept for 7 days, then deleted automatically. After that the Content section is empty. Copy anything you need before then. There is no setting to extend or switch this off. Payment and transaction records are kept separately. See Data retention.

Why an old charge can differ from the price on the models page

Model prices change over time. A charge is worked out when the request is billed, using the price in effect at that moment, and then taken from your wallet. A later price change affects new requests only. It does not change charges that were already made.

So if the models page shows a different price today than the one an older request seems to have used, that is expected after a price change. To compare, look at requests from the same day.

Checking a charge you do not understand

  1. Open the request and note its tokens: Prompt, Completion, Cached read and Cached write.
  2. Check whether the request used a different model than you expected. Some requests can be answered by another route for the same model, and the Overview shows the routing.
  3. Check the state. Was it completed, failed, or cancelled by your client?
  4. If the charge still looks wrong, write to support with the request time and model from the row. We can check it against the stored record.

Keep one key per project. The key budget stops one runaway script from using the whole wallet. See Authentication.