Expand description
Context usage of a chat and its compactions.
When a conversation nears the model’s context window, the engine asks the model for a dense summary and continues from that summary plus the newest turns (“compaction”). The transcript keeps every message; clients show a divider where a compaction happened.
Structs§
- Chat
Usage - Token use of one chat, for the header meter.
- Compaction
- One compaction of the chat in view.
Functions§
- short_
tokens - “950”, “9.4k”, “9k”, “34k”, “1.2M”.