langchain

Why we included this project

Elixir and Phoenix teams that want to talk to large language models from their existing BEAM applications will find a purpose-built fit here. It is a native Elixir take on the LangChain idea, so you add it as a library rather than standing up a separate Python or JavaScript service. Adapters cover a wide range of providers, including OpenAI, Anthropic Claude, Google Gemini, xAI Grok, Ollama for local models, and Bumblebee for fully self-hosted inference, with a generic ReqLLM adapter that reaches many more. Beyond plain chat calls, it provides the abstractions that make LLM features workable in production: chaining steps together, tool and function calling so a model can act on its environment, structured output, and hooks into external data sources. That makes it a practical choice whether you are adding a chat endpoint, building an agent that takes actions, or connecting a retrieval step into an existing Elixir service, and the building blocks stay consistent with the rest of your codebase.