← all repositories
ItzCrazyKns/Vane

Self-hosted answer engine that cites sources and never phones home

A privacy-focused AI answering engine that runs entirely on your own hardware so your searches never leave the building.

35.8k stars TypeScript RAG · SearchAgents
Vane
Velocity · 7d
+24
★ / day
Trend
accelerating
star history

What it does

Vane is a self-hosted AI answering engine built on Next.js that synthesizes answers from web search and uploaded documents while citing its sources. It routes queries through SearxNG to aggregate results from multiple search engines, then passes the gathered context to local LLMs via Ollama or to cloud providers such as OpenAI, Claude, and Groq. The application also handles file uploads, domain-specific searches, and maintains a local search history.

The interesting bit

The project is mostly well-organized glue between SearxNG, various LLM APIs, and a Next.js frontend. Its small flourishes—widgets that surface weather, stocks, or calculations when relevant, plus tiered search modes for speed versus depth—add the polish that turns a simple RAG pipeline into a daily search replacement.

Key highlights

  • Bundles its own SearxNG instance in the recommended Docker image, so web search works immediately without external search APIs
  • Supports both local models through Ollama and major cloud providers, letting you trade privacy for speed as needed
  • Offers tiered search modes (Speed, Balanced, Quality) and domain filtering for targeted research
  • Surfaces widgets for instant answers alongside traditional cited responses, and can be set as your browser’s default search engine
  • Exposes a REST API for developers who want to integrate its search pipeline into other applications

Caveats

  • The troubleshooting section is extensive: expect to debug Ollama networking, Lemonade endpoints, and SearxNG JSON settings before everything works smoothly
  • Several promised features remain unchecked, including Tavily/Exa search support, expanded widgets, and user authentication
  • Non-Docker installs require manually installing and configuring SearxNG first, including enabling Wolfram Alpha and JSON output

Verdict

Developers who want a cited, AI-assisted search experience on private hardware—and are comfortable reading Docker logs—will find Vane a capable foundation. Those seeking a polished, zero-config SaaS experience should wait or look elsewhere.

Frequently asked

What is ItzCrazyKns/Vane?
A privacy-focused AI answering engine that runs entirely on your own hardware so your searches never leave the building.
Is Vane open source?
Yes — ItzCrazyKns/Vane is open source, released under the MIT license.
What language is Vane written in?
ItzCrazyKns/Vane is primarily written in TypeScript.
How popular is Vane?
ItzCrazyKns/Vane has 35.8k stars on GitHub and is currently accelerating.
Where can I find Vane?
ItzCrazyKns/Vane is on GitHub at https://github.com/ItzCrazyKns/Vane.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.