Description
Vane is an open-source AI answering engine, published under the name Perplexica until its rename; the original perplexica.io site is down, but development continues on GitHub at ItzCrazyKns/Vane, and the old ItzCrazyKns/Perplexica repository now redirects there. It's built to run on the user's own hardware rather than as a hosted service, combining live web search with an LLM to produce sourced, cited answers.
Read More
Vane runs as a self-hosted app, most commonly via a single Docker container that bundles Vane with the SearxNG search engine it uses for web queries; users can also point it at their own existing SearxNG instance instead. On first run, a setup screen lets a user add API keys and choose which AI provider and model to use.
The project supports local LLMs through Ollama alongside cloud providers including OpenAI, Anthropic Claude, Google Gemini, and Groq, so models can be mixed depending on the task. Search runs in one of three modes: Speed for quick answers, Balanced for everyday use, or Quality for deeper research. A search can be restricted to specific sources (web, discussions, academic papers) or a specific domain, and users can upload files (PDFs, text, images) to ask questions about their contents, or search for images and video alongside text. The interface also surfaces small utility widgets, such as weather or stock prices, when relevant, and keeps a local search history.
Vane is aimed at developers and privacy-conscious users who want a self-hosted alternative to hosted AI search products, keeping queries off third-party servers. It's released under the MIT license and has been public since April 2024; at last check the repository had roughly 37,000 GitHub stars and over 4,000 forks, with the maintainer describing it as under active, continued development.
Some advertised integrations, such as Tavily and Exa search support, are listed as "coming soon" rather than shipped yet. As a self-hosted tool it requires the user to run and maintain their own instance, whether via Docker or built from source; there is no first-party hosted version to sign up for.
- CategoryAI Search Engines
- Pricing ModelOpen source
- DeploymentSelf-hosted
- PlatformsWeb
Vane's Core Features
Self-hosted AI answering engine that runs via Docker or from source, keeping searches private
Web search powered by a bundled or external SearxNG instance across multiple search engines
Support for local LLMs (via Ollama) alongside OpenAI, Anthropic Claude, Google Gemini, and Groq
Three search modes: Speed, Balanced, and Quality, trading speed for depth
Source selection: restrict a query to the web, discussions, academic papers, or a specific domain
File upload support for asking questions about PDFs, text files, and images
Image and video search alongside text results
Locally saved search history and quick-lookup widgets (weather, stock prices, calculations)
Getting Started with Vane
Install: Run the bundled Docker image (itzcrazykns1337/vane:latest), which includes Vane and SearxNG
Configure: Open the app locally and enter your AI provider API keys, or point it at Ollama, on the setup screen
Pick a mode: Choose Speed, Balanced, or Quality mode depending on how deep you want the search to go
Search: Ask a question, optionally restricting it to specific sources or a domain, and get a cited answer
Upload files: Add a PDF, text file, or image to ask questions about its contents
Vane's Use Cases
- Private web search
- Local-LLM research assistant
- Document Q&A
- Developer self-hosting
