Running a Whole RAG Agent Offline: LangGraph + Ollama + Embedded Qdrant (Zero API Keys)

Short summary

A step-by-step guide to building a fully offline RAG agent using Ollama for local LLMs, embedded Qdrant for vector storage, and LangGraph for orchestration. The author demonstrates provider-swappable architecture with real execution results, including an intermittent failure case where a 9B model occasionally returns empty responses. This pattern works entirely without API keys and scales from local development to production via configuration.

• •Build offline RAG agents with Ollama + Qdrant + LangGraph
• •Provider-swappable architecture (Ollama ↔ OpenAI, embedded ↔ remote Qdrant)
• •Real gotchas documented: intermittent empty responses from small local models

Generated with AI, which can make mistakes.