Vane

An answering engine that runs entirely on your own hardware, mixing web knowledge with local models and cloud providers, and citing the sources behind each answer.

Vane is an AI powered answering engine that runs entirely on your own hardware. Searches stay completely private as a result, and there is no server in the middle to log the query. Answers come with cited sources, so the engine's claims can be checked. The privacy claim depends on the local hardware doing the work, which is exactly why the README leads with it. The engine combines knowledge from the internet with support for local LLMs through Ollama and cloud providers like OpenAI, Claude, and Groq. TypeScript is the primary language and the license is MIT.

The README says Vane supports all major AI providers. You can use local LLMs through Ollama or connect to OpenAI, Anthropic Claude, Google Gemini, Groq, and more, and models can be mixed and matched based on needs. The mix and match flexibility means a local model can handle routine questions while a cloud model steps in for the hard ones. Three search modes cover the speed to depth spectrum: Speed Mode for quick answers, Balanced Mode for everyday searches, and Quality Mode for deep research.

Users pick which sources the engine draws on, including the web, discussions, or academic papers, with more sources and integrations in progress. Widgets are described as helpful UI cards that appear when relevant, such as weather, calculations, stock prices, and other quick lookups. More sources and integrations are listed as in progress, so the source list is expected to grow. They keep the answer pane useful beyond the answer itself.

• Source 1 · Project repository
• Source 2 · README