Help CenterFeaturesUsage & limits
Usage & limits
Features

Usage & limits

What it is

QU Studio uses agents you installed and signed in to yourself, so usage counts against your own Claude / ChatGPT account or API key. QU Studio charges nothing and pays for nothing on your behalf.

This page covers three things:

  • Seeing clearly: how much quota each agent has left, when it resets, whether it'll last at the current pace, and where it went.
  • Spending it where it counts: cheaper models for small jobs, strong ones for big jobs; background work doesn't compete with you for quota.
  • When it runs out: switch to the other agent and carry on, or wait for quota to come back and carry on automatically.

Quota: what a subscription account (like Claude Pro / Max, ChatGPT Plus) can use every 5 hours and every week. Once it's used up, you wait for it to reset. Pay-as-you-go API keys have no quota; you look at cost instead.

When to use it

  • You want to know how much work you can safely hand out today: look at the small ring at the bottom left, or "Limits" on the usage page.
  • You want to know where your money and quota went: "Cost" on the usage page.
  • One agent is nearly used up: switch to the other, or turn on "Carry on automatically when quota is back".

How to use it

A quick look any time

  • The small ring at the bottom left: shows the tightest quota. It turns yellow under 20% left, and red under 10%. Click it to list every quota for every agent. You can hide the ones you don't want to see, and with several computers connected you can switch which one you're looking at.
  • The small ring next to the input box: how much of this session's context (how much conversation the agent can keep in mind at once) is used. It turns yellow at 70% and red at 90%. Hover to see this turn's input, cache hits and output.
  • Type /usage in a session: check any time.

The usage page

Click usage at the bottom left to go in. You can view 7 days or 30 days. There are three tabs:

Limits: one section per agent, one card per quota.

  • In large type, the percentage left, when it resets and how long until then.
  • A vertical line on the bar is the "even-pace line": how much should be left by now at an average pace. If the bar is longer than the line, it'll last at this pace; if shorter, things will get tight later.
  • Below, how much you can still use per hour or per day.
  • At the bottom are the ones without quota: API keys, cloud providers, and ones not signed in, each with a line saying why. Ones not signed in have a button to copy the sign-in command.

Cost / Tokens (tokens are the unit models use to measure text):

  • A daily cost curve, one line per agent.
  • Broken down by agent, by project, by model, and by "who used it" (Q / working sessions / helpers).
  • "Spent in the background for you": how much went to incoming work, the watchdog (the background program that checks whether sessions are stuck), Try first, and Q's first look, in a separate column.
  • Cache hits and output are counted separately, so you can see how much the cache saved.

Cost is always converted at API prices, so things are easy to compare; subscription users aren't actually charged this. Codex only reports tokens; QU estimates a figure from public prices and marks it "estimate".

Each task also shows how many tokens each session used and how long it ran.

Model presets: picking the model and thinking effort for the job

A model preset is a combination of "which agent · which model · thinking effort · fast or not · permissions", each with a one-line When to use.

Four come to start with (only for agents installed on this computer):

PresetWhen to use
EverydayRegular work
Small jobsSmall changes like copy, styling, or adding tests; fast and light on quota
Hard jobsPlanning, big changes, problems with no obvious cause
CodexTrying another provider, or when Claude's quota is running low
  • Switch with one click when you open a session.
  • The one marked "default" is what new sessions start with.
  • When Q hands out work and sessions call in helpers, they pick by each preset's "When to use": "Small jobs" for copy changes, "Hard jobs" for big changes. A preset without a "When to use" is never picked automatically.
  • The task and the session header say something like "Q picked it by 'When to use': Small jobs".
  • The preset Q picks never has wider permissions than the default one.

(Change presets under "Settings › Agent › Model presets", up to 20.)

Where to change the model for each job

Used forWhere to change it
New sessionsSettings › Agent › Model presets: the one marked "default"
Q handing out work, sessions calling in helpersSame as above, picked by each preset's "When to use"
Q's main chatQ › Settings › Model › Default model
Discussion topicsQ › Settings › Model › Model for discussions
Tidying memorySettings › Memory & privacy › Model for tidying memory
Guessing your next message, naming tasksSettings › Agent › Model for next-line guesses and task names
HelpersIf not specified, the same as the session that called them in; for another, just tell the session

The first time you open the app, there's a screen that sets these up by use, with one click on "Use the recommended ones": light jobs like guessing your next message and naming things get a fast, cheap model.

Background work doesn't compete with you for quota

Everything QU does for you in the background stays under a quota line and backs off when quota is nearly full. Work you hand out yourself isn't affected:

  • The watchdog decides who's stuck by rules. It doesn't call a model or use quota.
  • Of incoming messages, ones from bots and ones that only say "+1" are filtered out first by fixed rules.
  • Incoming work and re-handed-out work isn't started once quota reaches 90%.
  • Try first stops at 80%.
  • The morning brief comes in a short version when quota is nearly gone; a scheduled task skips this run when less than 10% is left; tidying memory waits until there's plenty of quota.
  • Reading quota means asking the agent directly, which uses nothing.

When quota is nearly or completely used up

  • Nearly used up: when one agent has less than 20% left, the session asks "Continue on the other one?". One click switches it, and the task and the conversation so far come along. If the other one is also nearly out, not installed, or not signed in, you're told first.
  • Used up: the session stops at this turn, marked "Waiting for quota · back ~14:30". Changes are kept, not rolled back.
  • Carrying on automatically: turn on "Carry on automatically when quota is back", and it picks up by itself when the time comes, even after restarting the computer. (It's under "Settings › General".)
  • Q's own quota runs out: you can wait, or switch to the other agent and keep chatting, and switch back later.

Looking across several computers

  • The usage page can show "All combined / This computer / each computer": how much each computer spent and what share.
  • If several computers sign in to the same account, the quota is counted once, so it doesn't look like twice as much is left.
  • When handing out work, Q looks at which computer has a free slot and which agent has more quota left, and sends it there. Ones nearly full go to the back; ones used up get nothing.

FAQ

Does using QU use more quota? Q coordinating and looking work over first also uses a model. What QU can do is spend quota where it counts: light jobs use a fast, cheap model by default, and background work stays under the quota line. We have no comparison data on exactly how much it saves, so we don't give a number.

Is the money in the cost column charged by QU? No. It's a reference figure converted at API prices; QU charges nothing. Subscription users use quota as their subscription works.

Why do some agents have no quota card? API keys and pay-as-you-go cloud providers have no quota, only cost. Ones not signed in say why, and you can copy the sign-in command.

How often is quota data updated? It's reported by the agent in a session, so it only shows up after a turn has run.