LLM Agent Chat

Node.js agent chat with Weaviate RAG, Gemini LLM, and Chart.js tool. Supports streaming NDJSON output.

---

Setup

Prerequisites: Node.js 18+, Docker, Gemini API key

npm install
cp .env.example .env
# Edit .env and set GEMINI_API_KEY

Start Weaviate:

docker compose up -d

Wait ~10 seconds for Weaviate to be ready.

Seed the database:

npm run seed

Run the chat CLI:

npm run chat

Type your query, press Enter. Type exit to quit.

---

Example Queries

Type	Query	Expected
Direct	Explain what LangGraph does.	LLM answer, no RAG/chart
RAG	What does the safety guideline say about PPE for welding?	Answer with citations (1- Page 7, 1- Page 8)
Chart	Generate a bar chart configuration.	Chart.js config in data
Both	Based on the quality manual, summarize inspection sampling and return a chart config.	RAG answer + Chart.js config

Output is streamed as NDJSON, one JSON object per line: { "answer": "...", "data": [...] }.

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                    Delegating Agent                          │
│  (routes by keywords: direct | rag | chart | rag+chart)      │
└─────────────────────────────────────────────────────────────┘
         │              │                    │
         ▼              ▼                    ▼
    ┌────────┐   ┌─────────────┐   ┌──────────────────┐
    │  LLM   │   │ RAG Agent   │   │ Chart.js Tool    │
    │ Gemini │   │ (Weaviate)  │   │ (mock config)    │
    └────────┘   └─────────────┘   └──────────────────┘
                        │
                        ▼
              ┌─────────────────────┐
              │ Weaviate (Docker)   │
              │ QnAChunk, tenantA   │
              │ 3 seed documents    │
              └─────────────────────┘

• Delegating Agent: Parses the query, chooses direct answer, RAG, chart, or both.
• RAG Agent: Fetches chunks from Weaviate, formats references as N- Page X, returns grounded answer.
• Chart Tool: Returns a fixed Chart.js bar config (mock).
• Streaming: Each event has answer (text) and data (RAG refs and/or chart config).

---

Project Structure

app/src/
├── config/env.ts
├── weaviate/         # schema, seed, retrieval
├── llm/              # Gemini provider
├── tools/            # Chart.js mock
├── agents/           # delegating, RAG, graph
├── streaming/
└── cli/chat.ts

docker-compose.yml    # Weaviate
docs/                 # Implementation plan