Administration
Usage limits and context window
How the Agent's two limits work: usage (how much you can use in the period) and the context window (how much can fit in a conversation). They are different and independent systems — this is the most common doubt.
Concepts and terms
| Term (as it appears on screen) | What does it mean |
|---|---|
| Use | How much of the agent you used in the period. It works like a budget that is renewed every cycle and is shared by everything the agent does for you: conversations, announcements, reports, Routines. We never talk about tokens or messages — it's just one thing, "usage". |
| Context: N% (ring at the footer of the conversation) | How much of the context window does that conversation already occupy. The window is the agent's working memory of that conversation: everything that was said, the attachments and the data consulted. It's by conversation — opening a new conversation starts with an empty window. |
| "Organizing the conversation..." | Automatic compression: when the conversation gets long, the agent summarizes the beginning to continue with momentum, without losing the thread. It happens on its own — the user doesn't need to do anything. |
| Bar "Usage (30 days)" | Percentage of period usage, visible on the hover of the context ring and in the Plans modal. It only shows the percentage (e.g.: 25%), without internal numbers. |
How usage works
- Renews every period and covers everything the agent does for you — there are no separate counters by type of work.
- What consumes the most: very long conversations, large attachments and analyzes that cross a lot of data. Consumption is proportional to the work, not the number of messages.
- PRO Max is the same PRO plan with 5× greater usage — for those who use the agent as their entire routine. *(⚠️ Early access; hiring not yet available on the modal.)*
How the context window works
- Each conversation has its own. The Context: N% ring shows how much has already been used; It also appears on the home page.
- What occupies the window: the conversation history, attachments (up to 5 files per message; texts up to 120 thousand characters; images up to 2 MB) and the data that the agent consults to respond.
- When it fills up, the agent automatically compresses ("Organizing the conversation...") — summarizes the beginning and continues. The conversation doesn't stall or lose the topic.
- The window is not the usage limit: a conversation with 90% context does not mean that your usage for the period is high, and vice versa.
Tips to make the most of it
- Did you change the subject? Open a new conversation. Long conversations use more time for each response (the agent rereads everything) and fills the window faster.
- Use Projects for recurring materials — the agent searches for what they need from there instead of loading everything into each conversation.
- Only send necessary attachments — large files occupy the window and increase the consumption of that conversation.
What tends to confuse
- Use ≠ context window. Use is for the period and applies to everything; the window is from that conversation and resets to a new conversation.
- "How much can you use?" — the answer is in work equivalences ("a typical month of announcements, reports and day-to-day conversations"), never in tokens or number of messages. Tokens are an internal ruler, not product vocabulary.
- Compression does not erase the conversation — it summarizes the beginning for the agent; the visible history remains complete.
- Memory, Context (Personalize) and specialties do not compete with the conversation: the agent only injects what is relevant, instead of always loading everything. *(Plan doc principle; selective injection is an architectural item in V5.)*
Related Features
- Agent — the area, Conversations, Routines, Memory and Context