// the find
ItzCrazyKns/Vane
Vane is an AI-powered answering engine.
Vane is a self-hosted Perplexica fork — a Perplexity-style AI search engine that routes queries through SearxNG and feeds results to your choice of LLM (Ollama, OpenAI, Claude, Groq, etc.). It stores everything locally, so your search history stays on your machine. Best fit for developers who want a private, hackable alternative to Perplexity and are willing to run their own infrastructure.
The multi-provider LLM architecture is genuinely flexible — swapping between Ollama, Claude, and OpenAI is a settings change, not a code change. The researcher agent in `src/lib/agents/search/researcher/` follows a plan-then-act loop with distinct action types (web search, academic search, URL scrape), which is a cleaner design than the monolithic prompt approach most similar projects use. Docker setup is single-command and bundles SearxNG, so the 'getting started' path actually works. Drizzle ORM with migrations means the schema is version-controlled and reproducible.
No authentication — the README lists it as an upcoming feature, so anyone who can reach port 3000 owns your instance and search history. The widget system (weather, stocks, calculations) calls third-party APIs at query time, which quietly breaks the privacy pitch the README leads with. Local SQLite via Drizzle means there's no multi-user story; if you want to share an instance with a team, you're on your own. The codebase is a renamed fork of Perplexica with cosmetic changes, so upstream improvements won't land automatically and the project is carrying someone else's architecture decisions without clearly diverging from them yet.