Vane

Formerly known as Perplexica and renamed in March 2026, Vane is a privacy-focused answering engine that reproduces the Perplexity experience on hardware you control. It uses SearxNG as a metasearch layer, so queries fan out across search engines without exposing user identity, then an LLM composes answers with cited sources. Three modes trade depth for latency: Speed for quick answers, Balanced for everyday use, and Quality for deep research. Model choice is flexible, spanning local models through Ollama and hosted ones from OpenAI, Anthropic, Google Gemini, and Groq, mixable per task. The recommended install is a single Docker command that brings up the app, SearxNG, and persistent volumes together; a non-Docker path requires manual SearxNG setup and Node.js. An API allows embedding the engine in other applications, and architecture docs live in the repository. The MIT license permits commercial use, there is no paid tier, and no GPU is needed unless running local models. With 35.9k GitHub stars it is the leading open source Perplexity alternative.