srinivas191206/much

Much is a self-hosted AI chat workspace that brings together multiple cloud and local AI model providers, such as Gemini, Groq, OpenRouter, and locally run Ollama models, in one interface you can switch between mid conversation. Its main difference from similar self-hosted chat tools is where the heavier features run: instead of needing separate Docker containers for a code sandbox or a vector database, Much runs Python code and document search directly inside your web browser. The in-browser Python sandbox uses a project called Pyodide, which is Python compiled to run as WebAssembly, so you can run scripts, work with data tables through Pandas, and draw charts with Matplotlib without sending anything to a server. Document search works the same local way: when you upload PDFs, spreadsheets, or text files, Much reads them and calculates similarity between pieces of text right in the browser to answer questions about their contents, rather than relying on an external search service. Other features include generating and editing images with Stable Diffusion XL, which can run without an API key, or DALL-E 3 as a paid alternative, plus support for the Model Context Protocol so models can connect directly to filesystems, databases, and other tools. The interface can render generated React components, static HTML pages, and Mermaid diagrams live inside the chat. Sign in is handled through Google or GitHub accounts, and conversation history and credentials are encrypted locally and stored in MongoDB. To run it yourself you need Node.js version 18 or later and a local MongoDB instance. You clone the repository, install dependencies for both the frontend and the backend server folders, add environment variables such as a MongoDB connection string and OAuth credentials for Google and GitHub, then start the backend and frontend with separate npm commands. The project is released under the MIT license and can also be deployed to Railway or Render.