This change enables shipping Docker images with pre-indexed knowledge graphs,
eliminating the need to re-index documents in production deployments.
Benefits:
- Reduces embedding API costs (no re-indexing in production)
- Enables instant query capability (zero startup delay)
- Ensures consistent embeddings across deployments
- Simplifies multi-region and offline deployments
Changes:
- Modified Dockerfile to optionally copy pre-built graph file
- Updated .dockerignore to allow specific graph files through
- Added comprehensive rag_storage/README.md documentation
Implementation Details:
- Graph file: graph_chunk_entity_relation.graphml
- Copy is optional: builds succeed with or without the file
- .dockerignore pattern allows graph while excluding other storage
- Documentation covers use cases, best practices, and troubleshooting
Usage:
1. Build knowledge graph locally
2. Docker build automatically includes it if present
3. Deploy container with instant query capability
The feature is backward compatible - existing builds work unchanged.
• Replace pip with uv for dependencies
• Add offline extras to Dockerfile.offline
• Update UV_LOCK_GUIDE.md with new commands
• Improve build caching and performance
- Add Rust installation to Dockerfile for potential performance improvements
- Set explicit Docker data directories
- Add graspologic library to API requirements
- Optimize Docker build stages and environment setup
