opengpts

OpenGPTs is an open-source project that gives you your own version of what OpenAI calls GPTs and the Assistants API. Instead of being locked into OpenAI's choices, you get to pick the pieces yourself: which AI model to use (from over 60 options), what instructions to give it, which tools it can call on, and where to store the information it references.

The project is built on top of LangGraph, a framework for wiring together the steps an AI takes when responding to you. On top of that, it uses LangChain (a library of AI building blocks), LangServe (for running the AI as a web service), and optionally LangSmith (for debugging and logging). The combination means you can run the whole thing on your own computer or server rather than sending everything through OpenAI's infrastructure.

Three working configurations come included. The first, called Assistant, behaves like a general-purpose AI that can call tools such as web search or code execution. The second, called RAG, lets the AI answer questions by looking things up in a document collection you provide. The third is a basic chatbot without tool access. All three are written as regular code, so you can read them, copy them, and change them however you like.

Getting it running requires either Docker (which handles the setup automatically) or a manual install with Python 3.11 or newer, a Postgres database, and a few environment variables for your API keys. The front end is a browser-based chat interface that runs locally. You need at least one language model API key, such as an OpenAI key, to make the AI part work.

The project is openly licensed and encourages modification. If you want a self-hosted alternative to commercial AI assistant products and are comfortable with a technical setup process, this repository provides a working starting point.