Obelisk RAG Documentation¶

This document serves as the central reference for Obelisk's Retrieval Augmented Generation (RAG) implementation. It provides an overview of components, configuration options, and integration points, while directing readers to specific documentation files for detailed information.

Current Implementation¶

The Obelisk RAG system provides document indexing, vector storage, and query capabilities through a modular architecture with the following key components:

Core Components¶

APIs & Integration¶

User Guides¶

• getting-started.md - Step-by-step guide with RAG overview and first-time setup
• using-rag.md - Comprehensive usage documentation with examples
• implementation-status.md - Current implementation status and roadmap

Vector Database Implementation¶

The current implementation uses ChromaDB for vector storage. Future development roadmap includes potential migration to Milvus Lite for improved performance and scalability.

Documentation Progress¶

Based on the analysis in task.md, we've made significant improvements to the documentation:

Completed Tasks¶

✅ High Priority - Consolidated API to use only the OpenAI-compatible endpoint - Created comprehensive openwebui-integration.md documentation - Updated vector database documentation to reflect ChromaDB implementation - Fixed WebUI integration inconsistencies

✅ Medium Priority - Updated API documentation with OpenAI-compatible format - Fixed environment variable inconsistencies (OLLAMA_URL vs OLLAMA_API_URL) - Added detailed document chunking documentation to implementation.md - Documented error handling strategies and debug mode

All Tasks Completed ✅¶

1. Low Priority Items
2. ✅ Document the discrepancy between documented embedding batching and implementation
3. ✅ Update roadmap documentation to clearly mark implemented vs. planned features in architecture-draft.md
4. ✅ Add debug mode documentation
5. ✅ Improve troubleshooting guidance with more examples in using-rag.md

Documentation Maintenance Guidelines¶

When updating the RAG documentation, follow these principles:

1. Accuracy: Documentation must precisely match the implemented code
2. Completeness: Cover all aspects including setup, configuration, usage, and APIs
3. Examples: Include working code samples for key functionality
4. Future Direction: Clearly label future roadmap items vs. current implementation

Document Update Matrix¶

Related Documentation¶

Architecture documentation and detailed implementation information can be found in separate files:

• architecture-draft.md - Technical architecture with implementation status markers
• implementation-status.md - Implementation roadmap and current status

Implementation Notes¶

The current RAG implementation includes:

1. Document Processing
2. Markdown parsing with YAML frontmatter extraction
3. RecursiveCharacterTextSplitter for document chunking

4. Real-time file watching with the watchdog library

5. Embedding Generation

6. Ollama integration using mxbai-embed-large model
7. 1024-dimension vector embeddings

8. Support for both document and query embedding

9. Vector Storage

10. ChromaDB for vector database storage
11. Metadata filtering for search refinement

12. Local persistence on disk

13. Query Processing

14. Top-k similarity search with configurable parameters
15. Contextual prompt construction

16. LLM response generation via Ollama

17. API & Integration

18. OpenAI-compatible API endpoint for tool integration
19. Source information included in API responses
20. Docker containerization for deployment

Configuration Reference¶

The RAG system can be configured through environment variables: