felladrin/MiniSearch
Minimalist web-searching platform with an AI assistant that runs directly from your browser. Demo: https://felladrin-minisearch.hf.space
What it solves
MiniSearch provides a private, self-hosted search engine with an integrated AI assistant. It eliminates the need for third-party tracking, telemetry, and API keys by allowing the AI to run directly in the user's browser, while aggregating search results from multiple sources without exposing user queries to a single provider.
How it works
The system uses a Docker container to run an app server and a SearXNG metasearch instance. When a user searches, the server aggregates results and reranks them using a local cross-encoder model. The AI assistant then generates cited answers based on these results. This inference can happen in three ways: locally in the browser via WebGPU or CPU (using wllama), through a configured OpenAI-compatible API, or via the crowdsourced AI Horde.
Who it’s for
Users seeking a private alternative to mainstream search engines who want AI-powered summaries and cited answers without relying on on-device hardware or expensive API subscriptions.
Highlights
- In-browser AI: Supports 38 curated models (135M to 4B parameters) running on WebGPU or CPU.
- Privacy-first: No accounts, tracking, or telemetry; history and cached results are stored in the browser's IndexedDB.
- Flexible Inference: Supports local browser inference, remote OpenAI-compatible APIs, and AI Horde.
- Integrated Pipeline: Combines metasearch aggregation, local reranking, and cited AI responses in a single Docker container.
- Workflow Integration: Can be set as the browser's default search engine or triggered via Raycast.