Save tokens by selecting relevant outputs
under review
esaka
We can have option under each results to chose which results to reinject in discussion with LLM (to save tokens)
By default all previous results are send back to LLM, but the more we interact, we don’t want somes to pollute the discussion…
I know we already have option to send n previous messages, this one is more granular
VA
under review
J
Jpsala
Very useful, make a lot of sense!
M
Michal K
love this idea, it also makes the ai smarter due to less clutter
a
would also love to "sticky" messages to ensure they are sent even after context window
esaka
a, or "fork thread using only selected messages" :-)