FIGS: Figs Intelligent Graph Studio

Build Knowledge Graphs in One Command • Ground LLMs with Graph-Powered RAG

(Work in progress) but welcome stars & contributions!

FIG is an open-source toolkit that transforms your unstructured and structured data into actionable knowledge graphs, then serves them through a hallucination-resistant chatbot using graph-based Retrieval Augmented Generation (GraphRAG). Designed for developers who want:

✓ Precision over probability - Every answer grounded in your explicit knowledge graph
✓ Zero ETL graph construction - Start with tables (CSV/Excel), soon PDFs & databases
✓ Cypher-as-a-Service - Natural language to optimized graph queries via LLM translation

⚠️Early Access Alert
This is a proof-of-concept release (pre-alpha). Core features work, but expect rough edges.
Current focus: Table → Graph → Chat pipeline • Roadmap includes PDF/Parquet support

Quick Start (2-Minute Setup)

git clone https://github.com/yourusername/fig.git
cd fig
docker-compose up --build

Access the interface at http://localhost:3000.

Example Usage

Please refer to the README doc in data_gen for data generation and upload example.

Core Capabilities 🛠️

graph LR
    A[Sources] --> B(Schema)
    B --> C(Graph)
    C --> D{Playground}
    D --> E[Servers]

Loading

📊 Graph Studio: Visualize • Edit • Refine

• Live Graph Visualization: Explore nodes/relationships in an interactive canvas
• Precise Graph Editing: Directly modify node properties, relationship weights, and metadata

🧠 Schema Designer: Ontology Engineering Meets LLMs

• Drag-and-Drop Ontology Builder: Create node/relationship types using visual workflows
• AI-Assisted Schema Refinement: Chat with embedded LLM ("Should 'Customer' inherit from 'Organization'?")

📥 Data Sources Hub: Controlled Ingestion Pipeline

• Multi-Source Support: Allow structured, unstructure, static, and live data sources to be ingested into the graph.
• Precise Editing: Modify imported nodes/relationships in spreadsheet-like interfaces
• Source Tracking: Audit which source contributed each graph element.

🤖 Cypher Playground

• System Prompt Crafting: Guide LLM's Cypher generation ("Prioritize shortest-path queries") with a system prompt.
• Test before release: Side-by-side natural language ↔ generated Cypher comparison, execution time metrics, result previews before deploying as OpenAI compatible API.

🌐 Deployment Servers

• OpenAI-Compatible API: Drop-in replacement for existing chatbot applications(`/v1/chat/completions`)
• Raw Graph Connections: Bolt protocol support for Neo4j/Tigris direct access (WIP)
• MCP Server: Integrate with your own agent through Model Context Protocol. (WIP)

Roadmap: What's Cooking? 🔥

Data Source Ingestion:

• PDF Support: Extract structured entity & relationshipdata from PDFs.
• Database Integration: Connect directly to existing databases.
• Parquet Files: Handle large datasets efficiently.

Server Deployment:

• Raw Graph DB Connections: Access your graph without intermediaries.
• MCP Server: Add Model Context Protocol for better integration with agent ecosystems.

General Features:

• Model Selection for agents: Use your preferred LLM (Ollama, Gemini, Claude, Groq, etc.) for agents
• Import and export graph as RDF / OWL / JSON.

Contributing

We welcome contributions and feedback from the community! If you have ideas for improvements, bug fixes, or new features, please feel free to:

• Open an issue
• Fork the repository.
• Create a new branch for your feature or bug fix.
• Submit a pull request with a detailed description of your changes.