Skip to content

Latest commit

 

History

History
52 lines (40 loc) · 2.61 KB

USAGE.md

File metadata and controls

52 lines (40 loc) · 2.61 KB

AI usage

The AI usage section on the organization page shows where your AI requests, tokens and errors went, so you can see why you're hitting limits before you hit them. Everyone in the organization can see it. Nothing about prompts or replies is stored or shown: only counts, token numbers, model names and error types.

What you see

Item Meaning
Operations things DevPilot asked the AI to do (a review, an analysis, a question)
Provider calls requests actually sent to the provider, retries included. This is the number your quota counts
Failed operations operations that ended in an error
Avg response average time of the successful ones
Input / output tokens text sent and received. Output includes the model's hidden thinking
Last 24 hours provider calls in the last 24 hours, against your daily limit if you set one
Provider calls per day bars per UTC day; a red bar had failures
By feature Ask, code search, indexing, pull request review, CI analysis, connection test
By model calls per model. Provider daily limits are usually per model
What went wrong error types: daily quota used up, rate limited, provider unavailable, timeout, API key rejected, and so on

When the provider says a daily quota is used up, a banner appears for 24 hours with when it was last seen. DevPilot stops retrying at that point (docs/AI.md), so one bad day doesn't burn requests.

Show your limit

DevPilot can't read your quota from the provider, so you tell it. In apps/api/.env:

AI_DAILY_REQUEST_LIMIT=20

Use the number from your provider's rate-limits page (the Google AI Studio free tier is small). The Last 24 hours bar then fills up and turns red at the limit. Leave it unset to hide the bar. Restart the API after changing it.

Why the numbers may differ from your provider's

  • DevPilot counts only requests it sent. If the same API key is used by other apps, the provider counts those too.
  • Days here are UTC. A provider may reset its day at another time (some use midnight Pacific), so the 24-hour bar is a guide, not the exact quota state.
  • Retries count. One slow review that retries three times is three provider calls. That is usually why a small quota runs out faster than the operation count suggests.

Tips for a small quota

  • Look at By feature: reviews and indexing use the most calls.
  • Don't click Analyze or Review repeatedly while a request is queued.
  • Check By model: a different model has its own daily allowance.
  • Turn on billing with your provider for real use.