h2ogpt

h2oGPT is an open-source tool that lets you run an AI chat assistant entirely on your own computer or server, keeping your data private. You can have conversations with the AI and also upload your own documents so the AI can read, search, and answer questions about them. Supported file types include PDFs, Word documents, Excel spreadsheets, images, audio files, video frames, YouTube video transcripts, and plain text, among others.

The software supports a wide range of AI models, including LLaMa 2, Mistral, Falcon, and many others, as well as models accessed through tools like Ollama and llama.cpp. Depending on your computer's hardware, models can run on a graphics card for speed or on the CPU alone. The interface is a browser-based panel built with a tool called Gradio, where you can upload documents, start conversations, and manage multiple separate document collections for different projects.

Beyond basic chat, h2oGPT includes voice capabilities: you can speak to it using microphone input and have it respond in audio. It also supports image generation through several tools including Stable Diffusion, and it can describe the contents of images using vision-capable AI models. Web search integration lets the AI look up current information to include in its answers.

For developers, the project can act as a drop-in replacement for OpenAI's API, meaning existing applications built to talk to OpenAI's servers can be pointed at h2oGPT instead to run privately. It also supports agents that can write and execute code, analyze spreadsheet data, and produce charts.

Installation is available via Docker (the recommended method for most users), as well as platform-specific guides for Linux, macOS, and Windows. The project is licensed under the Apache 2.0 open-source license, meaning it is free to use and modify.