Vane is an open-source, privacy-focused AI answering engine designed to run on the user’s own hardware. It combines web search with local or cloud language models to produce answers with cited sources while keeping the search application, history, uploaded files, and optional local inference under the user’s control.

The system supports Ollama as well as providers such as OpenAI, Anthropic, Gemini, and Groq. Users can choose different search-quality modes, search the web, discussions, or academic sources, upload documents for question answering, search images and videos, restrict queries to specific domains, and use contextual widgets for quick information. Web retrieval is primarily powered by a bundled or external SearxNG instance.

Vane is useful for self-hosters and privacy-conscious researchers who want a Perplexity-style answering experience without depending entirely on a hosted search product. Docker deployment bundles the application with SearxNG for a relatively simple setup, while advanced users can connect their own search instance and mix local models with cloud models according to privacy, quality, and cost requirements.