Chat

The Chat tab lets you query the knowledge graph with natural language. Answers stream from GraphRAG’s query methods and highlight relevant entities in the visualization.

Modes

• drift: Context-aware hybrid retrieval with narrative output
• local: Focuses on local neighborhoods
• global: Aggregates across broader graph context
• basic: Minimal, fast baseline

Switch modes via the selector under the chat header.

How It Works (Stream)

• Endpoint: POST /api/chat/stream
• Sends Server-Sent Events (SSE) including:
  • status: lifecycle updates
  • step-start / step-end: pipeline phases
  • drift-log: raw tool output (sanitized)
  • highlights: entity IDs to highlight in the graph
  • answer-chunk / answer: the prose answer

The app derives highlights from the GraphRAG output and builds a compact context from entities.json and relationships.json to inform the final display.

How It Works (JSON)

• Endpoint: POST /api/chat
• Returns { answer, highlights } after running the selected GraphRAG method (non-streaming).

Cost Estimate

The UI shows a rough token cost estimate (OpenAI pricing) based on input length and a fixed output token budget.