AI Stack architecture enhanced: vector database scaling for 1,000+ documents
MAJOR CHANGES:
- Dual deployment: AnythingLLM (primary) + Open WebUI (staff)
- AnythingLLM with LanceDB vector database (proven to 5,000+ docs)
- Workspace-based isolation (Operations, Pokerole, Brainstorming, Staff)
- Batch ingestion strategy: 100 docs at a time (prevent OOM)
- Embedding model added: all-MiniLM-L6-v2 for semantic search
- Detailed ingestion process (6-8 hours total, 2-3 hours active)
- Hardware validation: TX1 can handle 1,000+ docs easily
SCALING STRATEGY:
- Phase 3 now includes proper RAG pipeline
- Vector DB for semantic search across full repo
- Workspace isolation prevents context bleed
- Auto-update via Git sync (repos update → workspaces re-ingest)
RESOURCE UPDATE:
- Total: 140GB RAM / 251GB = 56% usage (was 44%)
- Total: 179GB storage / 809GB = 22% usage (was 15%)
- Headroom for 5,000+ document growth
Based on research: AnythingLLM gold standard for document-heavy self-hosting
Updated by: Chronicler the Ninth
mkrause612
25892bc402