Quick answer
A handful of prompts can exhaust a Cursor allowance because an agent run makes many model calls and premium models are priced per call. Check three things in order: whether you used a premium model, whether the runs were long agents, and whether another device or teammate is using the same account.
Four causes, in order
- Premium model selected: each call costs far more than Auto.
- Long agent loops: one prompt becomes dozens of model calls.
- Large context: every call resends the repository context.
- Shared or stale session: another device is spending the same allowance.
Check each one
| Check | Where | What it tells you |
|---|---|---|
| Model per request | Request details | Premium or Auto |
| Request count per prompt | Dashboard usage | Agent loop length |
| Recent sessions | History | Repeated retries |
| Active devices | Account sessions | Someone else spending |
Fix it
- Switch the default to Auto and escalate only after a failure.
- Stop the agent when the same error repeats twice.
- Start a new chat per task instead of continuing a long thread.
- Sign out unused devices.
- Set a remaining-capacity alert so the next stop is visible in advance.
If the allowance is genuinely gone, the usage limit guide covers the recovery options.
The diagnostic order
| Check | What it rules out | Where |
|---|---|---|
| Model used per request | Premium drain | Request details |
| Calls per prompt | Agent loop length | Usage tab |
| Session length | Context growth | Editor history |
| Other devices | Shared spend | Account sessions |
Work down the list. Most reports of three messages emptying an allowance end at the first row, because a premium model was selected and the run made dozens of calls.
The limit fix guide covers what to do once the allowance is gone.
Questions, answered
Why does Cursor say I am out of requests after a few prompts?
Agent runs make many model calls, and premium models cost more per call. A few prompts can be dozens of billable requests.
Does Cursor count each message as one request?
Not necessarily. An agent turn can trigger multiple model calls, and premium models carry different rates.
Does restarting Cursor restore requests?
No. The allowance belongs to the account and billing period, not the editor process.