You decide what the model sees.
A context dial on every turn — what the model reads, and what you pay for.
one turn, dialled
attachclient-alpha KB · retrieval only
loadbrief-v2 · full text
skipweb search off this turn
✓the model sees exactly that
Why it's good
Context is a budget. Spend it on purpose.
Every turn, you choose what rides along: attach a knowledge base for retrieval, load a topic in full, or send nothing but the question. Sharper context means sharper answers — and you only pay for the tokens the turn actually needs.
Attach vs load
Retrieval-only when the KB should answer, full text when the model must read every word.
Scoped per chat
Each chat sees the knowledge you point it at — nothing else bleeds in.
Cheaper by design
Tokens are the bill. Sending less, better-chosen context is the discount.
Sharper answers
Less noise in, less noise out — the model reasons over what matters, not everything.