Keeping your experience in the answer

An answer is more useful when you can see what it was based on. CueFlect checks the size of each request and shows which saved material made it into the answer.

Make room for the essentials

A model's context window has to hold both the request and the answer. CueFlect reserves space for the answer first, then measures the composed request. The question, system instructions and instruction against inventing your experience are treated as required material.

Screenshot text is bounded so each screenshot stays represented. When conversation history needs to be shortened, older turns are removed first to keep recent follow-ups in context.

See what was included

The overlay's Why this answer? panel identifies the profile and story material used, along with anything omitted. If a story was left out, the answer did not use it. When a relevant story is unavailable, CueFlect can offer a labelled framework for you to complete.

Know when to adjust a setting

If the required parts still cannot fit, CueFlect does not send the request. The message shows the estimate, the space reserved for the answer and the context setting to review (Ollama:NumContext for the local model).

These checks help preserve the grounding instructions; they do not guarantee a correct answer. Review the result and speak from your own experience.

The first-session guide shows how to add the profile and stories that make this useful.