LangChain is a framework for developing applications powered by large language models (LLMs). The MIT-licensed ecosystem provides the whole stack for agentic applications in three layers: LangChain as the framework with integrations for models, tools and vector databases; LangGraph as the low-level runtime for durable, stateful agent workflows; and DeepAgents, which builds ready-to-run deep agents with planning, sub-agents and a virtual filesystem on top of both.

---

We use the different layers across our Python AI systems. MUCGPT implements its assistants as agents created with DeepAgents, running on LangGraph with agent state persisted in PostgreSQL. The agent-based search of RIS-KI is likewise driven by a LangGraph agent. The RAG pipelines of the Dienstleistungsfinder KI-Suche and Zammad-AI use the LangChain framework with its integrations, for example for Qdrant.

All layers connect to OpenAI-compatible model endpoints and come as Python libraries, so they are used inside our own services — we consume the ecosystem without operating additional LangChain infrastructure.