1 comment

[ 2.9 ms ] story [ 12.9 ms ] thread
Putting together all I know about tokenminning. Point your agent at it.

1. Prompt hygiene (schema over prose, make no mistake) 2. Cache whenever you can 3. Context hygiene (seriously, start a new chat, it doesn't take much) 4. Routing 5. Control max_tokens, be smart about RAG 6. Semantic caching

What works for you?