Skip to main content

What is RAG?

RAG (Retrieval-Augmented Generation) is a technique that enhances the responses of Large Language Models (LLMs) by retrieving relevant information from an external knowledge base and providing it as context for the generation process. It acts as a bridge between the static knowledge of the model and dynamic, specific information sources.

How It Works​

RAG operates in two main phases:

  1. Retrieval: When a query is received, the system searches a knowledge base (like a collection of project documents or code snippets) for the most relevant information. This is often done using semantic search over vector embeddings.
  2. Generation: The retrieved information is then combined with the original query and passed to the LLM. The model uses this augmented context to generate a more accurate, detailed, and context-aware response.

Think of RAG as giving the AI an "open-book" exam. Instead of relying solely on its memorized training data, it can look up specific facts and patterns from your project's documentation to answer questions and generate code.

Common Questions​

  • Is RAG a specific tool? RAG is an architectural pattern, not a single tool. Reasoning implements a built-in RAG server, but the concept can be applied with various vector databases and retrieval systems.

  • Does RAG replace fine-tuning? No, it complements it. Fine-tuning adapts the model's core behavior, while RAG provides specific, up-to-date information at runtime. They can be used together for optimal results.

  • How is the knowledge base created? You build the knowledge base by adding documents, code snippets, or any text-based information using the provided tools. This content is then converted into embeddings for efficient searching.

RAG in AI Cockpit Reasoning​

AI Cockpit Reasoning implements a built-in RAG server to:

  • Connect to a local, persistent knowledge base.
  • Provide a consistent interface for adding and retrieving project-specific context.
  • Extend the AI's knowledge without needing to retrain or fine-tune the model.
  • Enable highly contextualized code generation and analysis on demand.

RAG provides a powerful way to make the AI assistant aware of your project's unique conventions, architecture, and requirements, leading to significantly better results.

Learn More About RAG​

Ready to dig deeper? Check out these guides: