Nexarag: Democratizing Reproducible Knowledge Graph Contexts for LLM Research - Automated Testing and Documentation

Nexarag Automated Testing and Documentation

Release

• Version: 1.1.0
• Release type: Testing and documentation
• Scope: Extended automated test coverage and enhanced documentation for onboarding and contribution

Highlights

• Frontend and backend automated testing
• Detailed video guides for deployment and usage
• Contribution standards and steps
• Tightened paper language to conform to JOSS standards

- Jupyter Notebook
Published by ben-n-fuller 3 months ago

Nexarag: Democratizing Reproducible Knowledge Graph Contexts for LLM Research - Nexarag Initial Release

Nexarag Initial Release Notes

Release

• Version: 1.0.0
• Release type: Initial public release
• Scope: End-to-end local platform for research knowledge-graph construction, exploration, and AI-assisted querying

Highlights

• Build knowledge graphs from Semantic Scholar search results and BibTeX imports.
• Expand graph context by adding paper citations and references.
• Upload PDF/Markdown/Text documents as graph-linked artifacts; PDF uploads are converted to markdown for downstream embedding.
• Query graph context through integrated chat (RAG-style responses over papers, metadata, and uploaded documents).
• Generate embedding-based 2D visualizations (PCA) with label-based grouping.
• Export, import, list, switch, and delete saved knowledge graphs.
• Integrate external AI clients via MCP over HTTP (/mcp).

Included Components

• Frontend: Angular UI with interactive Cytoscape graph exploration and Plotly visualizations.
• API service: FastAPI endpoints for papers, graph operations, documents, chat, visualization requests, health checks, model listing, and KG management.
• KG worker service: asynchronous graph-building, embedding, and chat response workflows.
• MCP service: tools for paper search/addition and language-to-Cypher query execution.
• Data services: Neo4j (graph store), RabbitMQ (event bus), Ollama (local model runtime).

Deployment and Runtime

• Deployment model: Docker Compose (Mac, Linux/WSL CPU, Linux/WSL GPU profiles).
• Default local endpoints:
  • Frontend: http://localhost:5000 (or 5100 on Mac profile)
  • API: http://localhost:8000
  • MCP: http://localhost:9000/mcp
  • Ollama: http://localhost:11434
• Core runtime requirements:
  • Docker (plus WSL2 on Windows)
  • Ollama models for embedding/chat/MCP flows (defaults: nomic-embed-text:v1.5, gemma3:1b, qwen3:8b)

Configuration

• Environment-variable driven runtime configuration for ports, credentials, model selection, and embedding chunking.
• Graph export artifacts and metadata are persisted under kg_dumps (mapped Docker volume/path).

Known Limitations

• Semantic Scholar ingestion is rate-limited and can delay graph updates.
• Knowledge graph import/export operations are designed around one active graph context at a time.
• API CORS is currently permissive and service authentication is not enabled by default in this release.

Test Coverage

• Backend test suite present for key utility and messaging paths, including:
  • API upload and socket utility behavior
  • RabbitMQ message serialization/publish/connection logic
  • Semantic Scholar retry logic

- Jupyter Notebook
Published by ben-n-fuller 5 months ago