chatchat-space/langchain-chatchat

Langchain-Chatchat (formerly Langchain-ChatGLM) is a Python application that lets you run a private, offline AI assistant powered by locally hosted language models. The problem it addresses is that services like ChatGPT send your data to external servers. Langchain-Chatchat runs entirely on your own hardware, no internet connection required for inference, so sensitive documents never leave your machine. How it works: you host one of the supported open-source language models locally using a model-serving framework such as Ollama, Xinference, or FastChat. Langchain-Chatchat then connects to that model through the LangChain orchestration library and adds a document question-answering pipeline (called RAG, Retrieval-Augmented Generation). The pipeline works like this: documents are loaded, split into chunks, and converted into numerical vectors using an embedding model. When you ask a question, the question is also vectorized, and the system searches the document store for the most similar chunks. Those matching chunks are combined with your question into a prompt, which is sent to the language model to generate a grounded answer. The web interface is built with Streamlit, and a FastAPI service exposes the same functionality as an API. Beyond document chat, the app supports Agent mode, where the model can autonomously call tools such as web search, database queries, ArXiv paper lookup, or Wolfram Alpha, as well as multi-modal image conversations using models like Qwen-VL. You would use this when you need a ChatGPT-like assistant for internal documents (legal files, research papers, company knowledge bases) where confidentiality requires everything to stay on-premises, or when you want a free, controllable alternative to subscription AI services. The tech stack is Python 3.8, 3.11, with LangChain, FastAPI, Streamlit, and FAISS or other vector stores. Supported model frameworks include Ollama, Xinference, LocalAI, and FastChat. Docker deployment is available.