Expose cache status in /health (deploy verification + ops signal)

Redis caching (in-container) for chats/messages/documents/memories with invalidation

Add idempotent light migrations so new columns (storage_path, etc.) apply to existing Postgres tables

Supabase Storage for raw uploaded files (gated, local fallback) + signed download endpoint

Conversation summarization for long chats + inline citations in SSE metadata

Supabase migration: Postgres engine (psycopg2) + pgvector vector store with Chroma fallback

Backend: temperature + persona (system_prompt) in chat; DOCX/CSV document loaders

Memory overhaul: cross-session (client_id), LLM extraction+dedup, provenance, edit endpoint

Per-browser chat isolation: scope sessions by X-Client-Id so users can't see each other's chats

Escape literal braces in prompt context so doc/memory text with {} doesn't break ChatPromptTemplate

Switch chat to Groq (fast+free), embeddings to local CPU; drop Ollama from image

Install zstd for Ollama extraction

Nexus Memory backend on HF Spaces (Docker + Ollama)