MiniSearch: Self-hosted web-search platform with AI assistant in the browser (github.com)

MiniSearch is a minimalist, open-source self-hosted web-search app that pairs a browser-based AI assistant with privacy-first defaults: no tracking, no ads, and no data collection. It runs models directly in the browser (cross-platform on desktop and mobile), only loads and caches models on demand, and exposes a clean UI that you can wire into your browser address bar or Raycast for fast queries. A live demo is available and the project is published on GitHub with Docker images and easy build-from-source instructions for local deployment (default host: http://localhost:7860). For developers and teams, MiniSearch offers flexible execution modes and operational controls: run models client-side for maximum data privacy and lower latency, or switch to a remote OpenAI-compatible API by changing the “AI Processing Location.” You can also host an internal OpenAI-compatible endpoint and keep your API key secret via environment variables (INTERNAL_OPENAI_COMPATIBLE_API_*). Deployment is simplified with a ghcr.io Docker image and docker-compose examples; access can be restricted using an ACCESS_KEYS value in a .env file. Configurable search/generation settings, efficient model caching, and an invitation to contribute make this a lightweight, practical option for anyone wanting a private, customizable AI-powered search experience.