💡 Add advanced RAG techniques as built-in agent tools

[CharnaParkey]

Today OpenRAG's Agent can detect the intent of your question, write various kinds of queries with OpenSearch or use other tools as needed. The agent might use several different tools or loop a few times to get to an answer. This means the better the LLM you select for Tool Calling and Reasoning the better.

We're considering adding dedicated RAG techniques as agent tools that would provide out-of-the-box known advanced retrieval strategies. This would likely improve results for smaller models.

What this would include:

• HyDE (Hypothetical Document Embeddings)
• Multi-hop retrieval
• Query rewriting
• Other advanced retrieval patterns

Benefits:

• No need to implement these patterns from scratch
• Battle-tested implementations
• Easy to toggle on/off for experimentation
• Consistent performance across use cases

Questions for the community:

• Which specific RAG techniques would be most valuable to you?
• How important is customization vs. having sensible defaults?
• Would you want these as separate tools or as configuration options on existing retrieval?

Vote with 👍 if you'd use this feature!

Let us know your thoughts and use cases in the comments below.