AI API logs

Stockly → Settings → Diagnostics → AI API logs.

This log covers calls to the model provider only. Stockly keeps its own, separate log for the document extraction service, on its own screen. → API logs in Stockly

Note

The entry appears only for a role holding View and clear the AI call log (contains prompts and model responses). The rows carry what was sent and what came back, so the log stays outside the ordinary viewer role.

The list

Columns: Created, Provider, Type, Model, HTTP, Tokens (in / out), Duration, OK. Filters: Provider, Request type and Result — Success or Failure.

Open on a row shows the request and the response in full.

What it answers

What this costs. The token columns give the real figure per document — the simplest answer to “what does one invoice cost to analyse”.

Why it failed. The HTTP code and the response body separate a rejected key from a spent quota from a document the model would not take.

Why it was slow. The duration says whether the wait was the provider’s or the document’s size.

Switching it on

The log is off by default. Stockly → Settings → AI Assistant, the Config tab, holds Enable OpenAI API logs.

Clearing it

Retention (days) decides how long entries are kept. Without it the log grows with every document processed.

Rows can also go by hand: select them and use Delete (n). The confirmation is explicit — the record of the call disappears for good, while the call itself is unaffected. Same privilege as opening the log.