Call log
Every AI call Modbot made — feature, model, tokens, cost, time taken and what went wrong — and what the model was sent for the calls that flagged something.
Every time Modbot asks an AI model anything, it writes down what happened. Settings → AI → Call log is that list.
It is the place to look when a feature has gone quiet, when the bill is higher than you expected, or when you want to know exactly what the model saw before it flagged somebody.
What each row says
| Column | What it is |
|---|---|
| When | When the call was made |
| Feature | Moderation, Insights (which includes the sentence on an alert), Chat or Connection test |
| Model | The model that answered, marked Fallback when it was the second one |
| Outcome | See below, with the error underneath when there was one |
| In, Cached, Out | Tokens sent, tokens the provider had already cached, tokens written back |
| Cost | What the call cost, or a dash when the model has no price |
| Took | How long the provider took |
| Asked by | The person who pressed the button, when a person did |
Outcomes
| Outcome | What happened |
|---|---|
| Answered | The model answered. |
| Timed out | The provider did not answer within the feature's time limit. |
| Error | Modbot could not reach the provider, or what came back was not an answer it could read. |
| Refused | The provider answered with an error — a bad key, a model it will not run, a rate limit. |
| Limited | One of your own spend limits, or a team member's monthly allowance, stopped the call. Nothing was sent and nothing was charged. |
A call that did not answer and could have gone better on another model is tried once more on your fallback model. Both attempts get their own row, so the log always shows which model actually wrote the answer.
Filters
Filter by feature, by outcome, by model, and by a time range. Flagged only narrows the list to the calls that produced a flag. Show more loads the next hundred rows.
Seeing what the model was sent
Rows with an Open button keep the whole prompt and the whole answer. Those are:
- calls that produced a flag
- calls you started yourself: the Test button, Try it on AutoMod, Generate now on insights, and AI briefs, which are listed under Chat
Every other call keeps counts only. The prompt of a routine moderation check is somebody's profile text or Discord message, and keeping a copy of every check would build a second copy of your members that nobody asked for.
On the Flags page, a flag that came from an AI topic links straight to the call behind it.
Who can see it
| What | Permission |
|---|---|
| The list, with counts, models, costs and errors | See operational log |
| The prompt and the answer behind one call | Change settings |
They are different permissions because they are different things to show. The counts are Modbot's own record of what it did. The prompt and the answer are a member's own words — so reading them needs the same permission that decides where those words get sent in the first place.
How long rows are kept
Thirty days by default. Change it with Keep the call log for (days) on
Settings → AI → Connection; 0 keeps them for ever. Modbot deletes older rows once a day.
Where to next
Limits and spend
What AI costs by feature, the month-end estimate, money limits for everyone, a feature, a role or one person, a monthly allowance for each team member, and model prices.
Analytics
The Stats page (Growth, Activity, Moderation and Regulars), the VRChat and Discord pages' This week, the Bluesky page, and where the numbers come from.