relevance + diversity evals, batched scoring, math tooltips

Streaming candidate picker + side-index feedback loops

Session priors (P(bucket) + P(type)) + inline eval UI

Add MMR reranking with conversation-context query fusion

Route sub-intents to their own pools, rip out the LLM intent router

added deeper memories and new users.

dropped more bloat

Remove library bloat

Code refactor for extensibility