h2oGPT
Self-hosted AI chat and document Q&A runs on Linux, macOS and Windows with local or cloud models. Apache 2.0 licensed; archived and no longer maintained.
h2oGPT is a self-hosted ChatGPT alternative for people who want to chat with local models and ask questions about their own documents. The project is archived and no longer maintained. It's open source under Apache 2.0, with support for Linux, macOS, Windows and Docker.
With local models and its offline document database, chat and document processing can stay on your hardware. It also connects to cloud providers such as OpenAI, Anthropic, Google and Groq; requests sent to those services leave the local setup. Supported local backends include Ollama, llama.cpp, GPT4ALL and vLLM, with models such as Mistral, Mixtral and LLaMa2. It can use CPUs, CUDA GPUs and Apple Silicon.
Document Q&A covers PDFs, Word files, spreadsheets, code and Markdown, plus images, audio and video frames. You can keep personal or shared document collections, search their contents, and generate summaries or extract information across documents. OCR support helps it read text from scanned material.
The browser interface includes streaming responses and side-by-side model comparisons. Beyond text chat, h2oGPT supports image understanding with LLaVa, image generation with Stable Diffusion and Flux, and Whisper speech transcription. Spoken responses, voice cloning and hands-free voice control are also available.
For developers, an OpenAI-compatible API exposes chat, embeddings, audio and image generation to other applications. Open WebUI can use h2oGPT as its backend. API-based agents handle web research, document questions and Python code execution, including generating plots.
