Each chat tracks the tokens it spends and how much of the model's context window it uses. This page shows where to read those figures. To send messages and read answers, start with Ask a Mind.
A chip beside the message box carries two readings for the chat:
Both readings update while an answer is being written, and a ~ marks a value that is still an estimate. Click the chip to open the Conversation usage dialog. In a narrow panel or window, the tok/s part hides and the token count stays.
The dialog opens from the chip, as a popover on a desktop and a bottom sheet on a phone. It shows:
While a reply is streaming, the figures shown include a live estimate for it, and every estimate carries a ~. The dialog notes: "Values with ~ are estimated while the reply streams."
The ring beside the message box shows how much of the model's context window the chat has used. It stays hidden while the chat is under half full:
| Context window used | What the ring shows |
|---|---|
| Below 50% | Hidden |
| 50% to 74% | Shown, in its default colour |
| 75% to 89% | Warning colour |
| 90% or more | Destructive colour |
Hover the ring to read Token Usage: the tokens used out of the model's context window (used / max tokens), with the percentage to one decimal. For screen readers, the ring is announced as Context window {pct}% used.
The size of the window is the context length of the model the chat runs on. A model's context length is listed in Choose models and AI behaviour. A bar form of the same indicator uses lower thresholds: warning from 50% and destructive from 80%.
When the ring climbs, shorten the chat so the Mind has room to work:
/ to summarise the older turns and free up room.The commands and the Compact chip are described in Chat commands, shortcuts and privacy. What you see while the chat compacts is in Chat states and recovery.
It is the speed of the answer in tokens per second: how fast the model is writing. The reading changes while an answer streams.
The ~ marks a value that is still an estimate. While a reply streams, the figures include a live estimate for it; when the reply settles, the final figures replace the estimate.
The ring shows how much of the model's context window the chat has used. It stays hidden below 50% used, so it appears once the chat passes half of the window.
From 75% used the ring shows its warning colour, and from 90% it turns destructive. Hover the ring to read Token Usage with the exact counts.
The per-kind token rows are not shown for DataMind-hosted models. The rest of the dialog, such as the total and the timing figures, still appears.
No. Swarm agents run as separate chats, and the dialog notes: "Swarm agents run as separate chats, not counted above." Their tokens are listed in the Agents row instead.
In a narrow panel or window, the tok/s part hides and the token count stays.
The chip appears once the chat has usage figures to show. Until then, no chip appears beside the message box.
Run / to summarise the older turns and free up room, or turn on Auto Compact so the chat is compacted when needed. See Chat commands, shortcuts and privacy.
The chip counts the tokens the chat has spent. The ring shows how much of the model's context window is in use at that moment.