LangChain RAG Patterns: Agent vs. Chain

A modular RAG system comparing two LangChain patterns: Agent (autonomously decides when retrieval is needed) and Chain (always retrieves before answering). Supports multiple embedding and LLM providers with a pluggable architecture.

Agent vs Chain

• Agent pattern — the LLM decides autonomously whether to retrieve context, useful when queries may not need RAG
• Chain pattern — always retrieves relevant documents before generating, predictable and consistent
• Chain with sources — returns source documents alongside the answer for traceability

Multi-Provider Support

• Embeddings: HuggingFace (local), OpenAI, Cohere, Google Gemini
• LLMs: Ollama (local), OpenAI, Cohere, Google Gemini
• Document loaders: PDFs and web scraping