How limits keep things fair

Guest / Trial Mode

Daily limits by plan

How credits and limits work together

What does a credit cost?

Automatic actions in chat

In normal chat the assistant can decide by itself that your request is better answered with a picture, spoken audio, a short video or a real document, and produce it without you switching modes — for example “draw me a pie chart of these numbers” or “make me a deck with a cover image”.

These actions cost credits and count towards the same daily limits as the ones you start yourself. An image the assistant decides to generate draws on your daily image allowance exactly as one you request in Draw mode does; the same is true for audio, video and documents. There is no separate allowance and no discount for automatic actions.

Because you did not ask for them in so many words, each one is labelled where it appears with what was made, which model made it, and what it cost. If you would rather the assistant never did this, you can switch it off entirely — or per type — under Settings → Automatic Actions. Switching to Draw, Video or Audio mode yourself always works regardless of that setting, and your administrator may also have some types switched off for everyone.

Projects

A project is a workspace: one set of instructions, one shared memory and one library of files, used automatically by every chat inside it. How many projects you can have, and how many files each one may hold, depends on your plan — the table above lists both.

A project’s context is sent with every message you write inside it, and that costs credits on every turn, whether or not it turned out to be relevant to what you asked. This is the one thing worth understanding before you write a very long brief: a project that adds 4,000 tokens of instructions and retrieved files to each message costs meaningfully more per message than a plain chat.

Two things keep that in hand. Each plan has a ceiling on how much a project may add to a single message, and your project page shows how close your instructions and notes are to it. And when the ceiling is reached, the project’s own instructions are kept first — project memory, notes and retrieved file excerpts are trimmed before the brief is, so the most important part always survives.

Knowledge files are stored so the assistant can search them, which means they are held on our servers as readable text. If you use end-to-end encrypted chats, note that a project’s instructions, memory and knowledge are not covered by that encryption — they are project data, not message data. Keep anything you need encrypted out of a project’s knowledge base.

Modules

A module is a guided template: you answer a few questions and it assembles the prompt for you. Some run a single step; others chain several steps together, each one a separate call to the AI.

A multi-step module costs the sum of its steps, not one message. A pack that researches a topic, drafts an outline and then writes three sections is four or five billable calls, and a step that generates an image or audio is priced like generating that media directly. The module shows an estimated credit cost before you run it — based on real runs of that module where we have them, and an estimate where we do not.

Runs are bounded so a chain cannot spiral: there is a hard ceiling on how many steps a module may contain, on how many parallel variations a single step may produce, on total runtime, and on the credits any one run may consume. A run that hits a limit stops and tells you, rather than continuing to spend.

Which modules you can use depends on your plan, and some are restricted to specific plans. Modules that touch regulated subjects — legal, financial, medical, career and safety topics — carry a notice before you run them and a caveat on the result. They produce general information, not professional advice; see the Terms of Service for what that means in practice.

Memory

Remembered facts are sent with your messages, which means they cost credits on every turn in the same way a project’s context does. A large memory makes every message slightly more expensive, so the recall setting in Account → Memory is worth knowing about: sending only what is relevant to the current message costs less than sending everything you have ever saved.

Automatic memory extraction is itself a small AI call, billed to your account like a normal message. It is throttled rather than run on every turn, and it does not run at all for chats where the memory switch is off, or when auto-memory is disabled in your preferences.

Sharing & collaboration

A read-only share link costs nothing to view — nobody spends credits reading a shared conversation, and viewers do not need an account.

A collaborator sending messages is different. Their usage is billed to the account that owns the conversation, and it counts against that account’s daily limits and credit balance. If you invite several people into a busy project, expect the owning account to carry all of it. Collaborator access can be removed at any time from the chat or project it was granted on.

Chrome extension

The extension is the same account, the same limits and the same credit balance as the website. It is not a separate allowance: a message sent from the side panel counts against your plan exactly as one sent from the web app, and daily caps are shared between the two.

Page content you pull into the panel is sent as part of your message and billed as input tokens, so grabbing a very long page costs more than a short question. The extension signs in with its own revocable token, which you can withdraw at any time from your account without changing your password.

What counts as fair use

Generated files are not permanent storage

Files you upload to chat