The model behind Crunch can only read so much at once — call it its reading room. A long conversation, with many questions and answers, slowly fills that room. Crunch shows you how full it is, and when it gets full, it makes space so you can keep going in the same conversation.
Watch the meter
The pill beside the send button shows how much of the model’s reading room the conversation uses (Context used). It turns amber from 70% and red from 90%.
Compacting
When the room fills, Crunch compacts the conversation: from then on, the model reads a summary of the earlier turns — your questions, its answers and the datasets they referred to — and the last turns in full.
- It happens on its own at 70%.
- To do it yourself, click Compact now, or type:
/compact
Your transcript on screen never changes. Compacting only changes what the model reads, not what you see.
Good to know
- Tip: If a message says the conversation has grown past what the model can read, type
/compactand ask your question again.
Related
- Ask Crunch about your data — asking and reading answers.
- Choosing a model — pick the model for your next message.