AI usage log: calls, tokens, cost
How to see when, for which task, with which model and at what cost the AI was called, how to find failed calls, and how to delete old log rows.
The AI usage log keeps one row for every AI call: time, task (scope), provider, model and tier, tokens used, estimated cost, the answer and any error. Summary cards at the top show the volume.
Steps
- Open Master datas (Google / AI / WIP) → AI usage.
- Filter with Scope, Provider, Email ID, AI action, Date from and Date to; for errors switch on Failed calls only. Click Filter.
- The cards show the number of calls, the input and output tokens, the cost, the average response time and the number of errors.
- Open the full request and answer of a call with the eye icon of the row (Details).
Deleting log rows
- The Delete button removes the selected rows permanently.
- The broom button Delete old deletes all rows older than X days at once; 30 days by default.
The log rows are written by the functions that call the AI; rows cannot be added by hand.
Example
The accountant notices that an e-mail was assigned to the wrong company. In the AI usage log she enters the message's identifier into Email ID and opens the call details: the model decided with low confidence and the fallback model gave the same answer. Based on this she refines the extra context in the AI configuration.