GraphRAG Visualizer

Overview

GraphRAG Visualizer is a web application for visualizing and searching Microsoft GraphRAG artifacts. It provides graph visualization, data tables, and integrated search (Local, Global, DRIFT, Basic) — all driven by a unified config.yaml.

Originally created by Yan-Ying Liao. This fork adds multi-datasource support, a unified configuration system, and a custom FastAPI backend.

Features

• Graph Visualization: 2D / 3D interactive graph view.
• Data Tables: Browse parquet data in tabular form.
• Statistics: Overview of data totals and type breakdowns for entities, relationships, and communities. Click any type to see all items, click an item to view details.
• Community Explorer: Hierarchical drill-down graph of communities. Navigate the community tree from top-level down to individual entities and text units. Single-click a community to view its report; double-click to drill into children. Includes label toggle (Auto/On/Off) and force-directed layout with isolated node centering.
• Integrated Search: Local, Global, DRIFT, and Basic search via the built-in API server.
• Multi-Datasource: Configure multiple GraphRAG output directories in config.yaml and switch between them at runtime.
• Unified Configuration: Single config.yaml for server ports, model endpoints, search parameters, and data sources — no .env files needed.
• Local Processing: All data stays on your machine.

Quick Start

# 1. Edit config.yaml
vim config.yaml

# 2. Install Python dependencies for the API
pip install -r graphrag-api/requirements.txt

# 3. Install frontend dependencies
npm install

# 4. Start both API and frontend
./start.sh

• Frontend: http://localhost:16888
• API: http://localhost:16889

Configuration (config.yaml)

server:
  host: "0.0.0.0"
  frontend_port: 16888
  api_port: 16889

models:
  embedding_model: "Qwen/Qwen3-Embedding-8B-Alt"
  chat_model: "Qwen/Qwen3-Coder-480B-A35B-Instruct-FP8"
  api_base: "http://10.210.156.69:8633"

search:
  community_level: 2
  response_type: "Single Paragraph"

data_sources:
  - name: "graphrag-example"
    path: "/path/to/graphrag/output/"
    description: "My GraphRAG Dataset"

Data Sources

Add entries under data_sources. Each entry needs:

• name: Unique identifier
• path: Absolute path to the GraphRAG output directory (containing parquet files)
• description: (optional) Human-readable label

The first entry is loaded by default on startup. Switch between sources via the UI or the /api/datasources/switch endpoint.

Model Configuration

The models section configures the LLM and embedding models used for search:

• api_base: OpenAI-compatible API endpoint
• chat_model: Model for search queries
• embedding_model: Model for embeddings

Startup & Shutdown

./start.sh                  # Start both API and frontend (foreground)
./start.sh --daemon         # Start in background
./start.sh --api-only       # Start only the API server
./start.sh --frontend-only  # Start only the frontend

./stop.sh                   # Stop all services
./stop.sh --api-only        # Stop only the API
./stop.sh --frontend-only   # Stop only the frontend

Architecture

┌─────────────────────┐     ┌──────────────────────┐
│  React Frontend     │────▶│  GraphRAG API         │
│  (port 16888)       │     │  (port 16889)         │
│                     │     │                       │
│  - Data Source      │     │  - /api/datasources   │
│    Selector         │     │  - /api/parquet/*     │
│  - Graph Viewer     │     │  - /search/local      │
│  - Data Tables      │     │  - /search/global     │
│  - Statistics       │     │  - /search/drift      │
│  - Community        │     │  - /search/basic      │
│    Explorer         │     │                       │
│  - Search UI        │     │                       │
└─────────────────────┘     └──────────────────────┘
                                      │
                              ┌───────┴───────┐
                              │  config.yaml  │
                              │  (shared)     │
                              └───────────────┘

API Endpoints

Endpoint	Method	Description
/api/datasources	GET	List all configured data sources
/api/datasources/switch?name=xxx	POST	Switch active data source
/api/datasources/current	GET	Get current active data source
/api/parquet	GET	List parquet files in active source
/api/parquet/{filename}	GET	Download a parquet file
/search/local?query=xxx	GET	Local search
/search/global?query=xxx	GET	Global search
/search/drift?query=xxx	GET	DRIFT search
/search/basic?query=xxx	GET	Basic search
/status	GET	Server health check
/api/config	GET	Get server configuration

Helper Scripts

load_output.sh

Creates a symlink from public/artifacts to a GraphRAG output directory for development mode auto-loading:

./load_output.sh /path/to/graphrag/output

Graph Data Model

Derived from the GraphRAG Neo4j Cypher notebook.

Nodes

Node	Type
Document	RAW_DOCUMENT
Text Unit	CHUNK
Community	COMMUNITY
Finding	FINDING
Covariate	COVARIATE
Entity	Varies

Relationships

Source	Relationship	Target
Entity	RELATED	Entity
Text Unit	PART_OF	Document
Text Unit	HAS_ENTITY	Entity
Text Unit	HAS_COVARIATE	Covariate
Community	HAS_FINDING	Finding
Entity	IN_COMMUNITY	Community

Supported Parquet File Formats

• GraphRAG v2.x.x: entities.parquet, relationships.parquet, documents.parquet, text_units.parquet, communities.parquet, community_reports.parquet, covariates.parquet
• GraphRAG v1.x.x: Same names with create_final_ prefix

License

MIT License. See LICENSE.

Original project by Yan-Ying Liao.