Usage limits
Last updated: 2026-09-07
Every AI feature in Persian makes a paid model call. Limits keep one account, one bug or one runaway script from spending the budget that everyone shares. This page explains what counts, where the lines are, and what Persian does when you cross one.
1. What counts as an operation
Every time Persian asks an AI model to do something for you, that is one operation. Generating a PR overview, explaining a chunk of diff, drafting a review comment, briefing an issue: one each. Summarising the files in a pull request counts as one operation for the whole batch, not one per file.
Results are cached against the commit they were generated for. Opening the same PR again does not spend another operation unless the code changed.
2. The limits on Pro
| Limit | Pro | Why it exists |
|---|---|---|
| AI operations per day | 50 | Bounds how much AI work one account can run in a day. |
| AI reviews of the same pull request per day | 8 | Stops the same PR being re-reviewed over and over. |
| Files in one AI review | 75 | Keeps a single review from swallowing an enormous PR in one go. |
| Diff lines in one AI review | 5,000 | Keeps giant diffs from eating the whole budget. |
| Premium-model calls per day | 10 | The bigger model costs many times more per call, so it has its own cap. |
The daily counter resets at midnight UTC. The weekly figure in the usage popover is informational: it is seven days of the daily limit, not a separate cap.
3. What happens when you hit one
Persian prefers a smaller answer over no answer. When a limit trips, a dialog explains which one and offers the closest thing it can still do:
- Daily operations. Nothing more today. The dialog says when the counter resets.
- Too many files. "Review fewer files" runs the review on the files that fit.
- Diff too large. "Use summary mode" summarises the changes instead of going line by line.
- Premium calls used up. Persian switches to the faster, cheaper model on its own. You get a result, just a lighter one.
- Context fetching. "Skip context fetching" answers from the diff alone without pulling in related files.
A hard cap always wins: summary mode or fewer files will not get you past the daily limit.
4. Which models run
Persian routes each task to the model that fits it. Quick jobs such as triage, summaries and risk scores go to a fast Anthropic model. Heavier jobs such as drafting review comments and explaining a whole PR go to a larger one. Very long hunk explanations are upgraded only when the input is big enough to need it. If Anthropic is unavailable, Persian retries with OpenAI.
5. Where to see your usage
The usage meter in the header shows today and this week. The Usage tab in Settings shows the same figures with history. Both read from the same log that enforces the limits, so what you see is what counts.
Need more than Pro allows? Email team@persian.sh and say what you are running into.