Model limits, pauses and switching

A run never fails because one model ran out. It pauses, keeps everything finished, and lets you choose.

Your model's limits

On the model step, Advanced → Your model's limits:

1. Start from a preset (for example Gemini API (Google AI Studio), free tier: 5 requests a minute and 20 a day for gemini-3.8-flash), or type your own numbers.
2. Fields: requests per minute, requests per day, input and output tokens per minute, and how many at the same time.
3. See your real limits opens your provider's console: it is always authoritative, and limits change.
4. Check today's usage shows Used via RepoLantern today: what RepoLantern counted for this key and model (your own use elsewhere isn't included).

RepoLantern paces itself to stay under these limits. They apply to your key and model across all your guides.

When a limit runs out

The guide shows Paused: model limit with what ran out (for example the daily request limit of gemini-3.8-flash), when it resets, files read and sections written so far, and the money spent. Nothing finished is redone or paid for twice. You have three choices:

1. Wait for the reset. The page says when it resets, in time left and on the clock (for example in 3 min, about 12:18; the weekday too when it is more than a day away), and says so when it has passed. Tick Resume automatically when the limit resets and keep the tab open, or Add a reminder to my calendar and resume by hand later. When the reset is less than two minutes away, Resume automatically is the main button; once it has passed, Try again now is. Your key is never stored, so the tab must stay open for an automatic resume; after a reload the page asks for your provider's details again before you can resume.
2. Switch model or provider (below). It is the main button while the reset is further away.
3. Finish with what's done: press it, then Make the guide from what's done. You get a partial guide that lists what's missing, with no more model calls.

When a guide stops or is cancelled

If a guide stops for another reason, the page says It stopped: with the reason, and what's finished is kept:

• Resume continues from where it stopped.
• Run again from the start starts over; file summaries already made are reused, not paid for again.
• Switch model or provider and Finish with what's done work as above.
• Delete this guide removes it.

A cancelled guide (Cancelled. What was finished is kept.) can only be run again from the start or deleted.

Switch model or provider

1. On the pause screen press Switch model or provider.
2. Choose the provider and model, and enter what it needs (setup).
3. Press See what's left: remaining files and sections, model calls, the likely cost on the new model, and warnings (for example if its daily limit or your spending limit won't cover it).
4. If this provider or model hasn't seen your code yet, tick the consent box.
5. Press Continue on the new model. Finished work is kept; only what's left runs.

The finished guide says which models wrote it, and the cost report splits the spend by provider and model.

Fallback models

Before you start, Advanced → If the model runs out → Add a fallback model (up to three). Each has its own provider, model, keys and an optional spending cap. When one runs out, RepoLantern continues with the next without pausing. The review step names every fallback in the consent.