The Oracle

Sci‑fi RAG console powered by local Ollama models. Switch between multiple GGUF quantizations, retrieve context from a prebuilt HNSW index, and chat with a cursed starship Oracle.

Features

• Switch models at runtime via dropdown (no page reload)
• RAG over local HNSW index (cached retriever)
• Streaming responses with live Oracle bubble
• Per‑reply latency display; model switch divider entries
• Status endpoint indicating model name and RAG state

Requirements

• Node 18+
• Ollama running locally at http://localhost:11434
• Pulled models (examples not in this repo):
  • gemma-2-2b-it-Q4_K_M
  • gemma-3-1b-it-Q3_K_L
  • phi3-mini-q3kl

Install

npm install

Dev

npm run dev
# A warmup helper will hit / and /api/status automatically

Open http://localhost:3000.

Model Switching

• The header dropdown lists available models (from Oracle_Config.ts: AVAILABLE_CHAT_MODELS).
• Selecting a model updates the server‑side registry and subsequent chats use it.
• A divider appears in chat: SWITCHED TO <model> in <ms>.

RAG Index

• Index path: rag_data/ (see FAISS_PATH)
• Retriever is warmed on first access and cached

API Overview

• GET /api/status → { model, rag, availableModels }
• POST /api/model { model } → switch active model
• POST /api/chat → streams Oracle response

Configuration

Edit Oracle_Config.ts for:

• AVAILABLE_CHAT_MODELS
• DEFAULT_CHAT_MODEL_NAME
• EMBEDDING_MODEL_NAME
• OLLAMA_BASE_URL

Notes

• Scrollbars and dropdown styled for dark sci‑fi theme