About Vane 1.12.2 (formerly Perplexica) — Self-Hosted AI Answer Engine
Vane, formerly called Perplexica, is a self-hosted AI answering engine. You ask a question, it searches the web through SearxNG, reads what it finds, and answers with the sources cited, much like Perplexity, but running on your own machine. The model can be local, through Ollama or LM Studio, or hosted, such as OpenAI, Anthropic Claude, Google Gemini or Groq. It is MIT licensed and developed by ItzCrazyKns.
This page links to GitHub’s source archive of the v1.12.2 tag. Bineret did not build Vane and is not affiliated with it.
Perplexica is now Vane
Same project, new name. The GitHub repository moved to ItzCrazyKns/Vane, old Perplexica links redirect there, and the Docker image is now itzcrazykns1337/vane. If you are following an older Perplexica guide, the steps still apply with the new names.
What it does
- Three search modes. Speed for quick answers, Balanced for everyday questions, Quality for deeper research.
- Choose where to look. The web, discussions, or academic papers, or limit a search to specific domains.
- Widgets. Cards for things like weather, calculations and stock prices when a question calls for them.
- Images and video. Visual results alongside the text answer.
- Your files. Upload PDFs, text files and images and ask about them.
- Discover. A feed of articles and trending content to browse.
- History. Searches are saved locally so you can go back to them.
How private is it, exactly
Searches go through SearxNG, a metasearch engine that queries several search engines for you. Those engines see a request from the machine running Vane, not your browser, and there is no account involved. Two things still leave your machine: the search queries SearxNG sends out, and, if you use a hosted model, your question and the gathered results, which go to that model’s provider. With a local model through Ollama, only the searches do.
What changed in 1.12
1.12.0, in December 2025, was a rewrite. LangChain was removed in favour of the project’s own provider system, with function calling replacing XML parsing. 1.12.1 added LM Studio as a provider. 1.12.2, on 10 April 2026, reworked Deep Research into a repeating reason, search, scrape and extract cycle, added a Chromium-based scraper for modern pages, used embeddings to filter results before they fill the context window, and added timeouts so a search cannot hang.
Which version you actually get
1.12.2 is the newest tagged release, and it is what this archive contains. Development continued afterwards on the main branch, and the Docker latest image was last rebuilt on 1 September 2026. So docker run with latest gives you newer code than this download. Pick the archive when you want exactly 1.12.2, and Docker when you want current.
Requirements
Docker is the recommended route: one container, with SearxNG bundled, serving on port 3000. Without Docker you need Node.js and a SearxNG instance of your own with JSON output and the Wolfram Alpha engine enabled. Either way you need a model, local or hosted.
When to use something else
If you want to chat with your own documents rather than search the web, AnythingLLM is built for that. If a team needs one shared chat app with accounts, LibreChat fits better. And if you do not want to run anything, a hosted answer engine is less work.
Key capabilities
Answers with sources
Vane searches, reads the results and writes an answer with citations you can open. Quality mode's Deep Research repeats a reason, search, scrape and extract cycle before it answers.
Search without an account
Queries go through SearxNG, which asks several search engines on your behalf. The engines see your server, not your browser. With a hosted model, your question still goes to that provider.
Local or hosted models
Run a local model through Ollama or LM Studio, or use OpenAI, Anthropic Claude, Google Gemini or Groq. Since 1.12.0 the project uses its own provider system with function calling instead of LangChain.
More than web pages
Search discussions or academic papers, restrict a search to specific domains, see images and videos, and ask questions about PDFs, text files and images you upload.
Perplexica, renamed
The project is now called Vane. The repository moved to ItzCrazyKns/Vane with redirects in place, and the Docker image is itzcrazykns1337/vane. Older Perplexica guides still work with the new names.
Tag versus Docker
1.12.2, from 10 April 2026, is the newest tag and what this archive holds. The Docker latest image is rebuilt from main and was updated on 1 September 2026, so it runs newer code.
Questions & Answers
Is Vane the same as Perplexica?
Yes. The project was renamed. The repository is now ItzCrazyKns/Vane, old links redirect, and the Docker image is itzcrazykns1337/vane.
Is it free?
Yes, MIT licensed. Hosted model providers bill you for what you use; a local model through Ollama costs nothing beyond your hardware.
Does it keep my searches private?
Mostly. Search engines see requests from the machine running Vane rather than your browser, and there is no account. Your queries still go out to those engines, and with a hosted model your question and results go to that provider.
Why does Docker give me something newer than 1.12.2?
The latest image is rebuilt from the main branch and was last updated on 1 September 2026, after the 1.12.2 tag. Build the image from this archive if you need exactly 1.12.2.
Can it use my own SearxNG?
Yes, with the slim-latest image and SEARXNG_API_URL. Enable JSON output and the Wolfram Alpha engine in SearxNG first.
Can I ask about my own files?
Yes, upload PDFs, text files or images and ask questions about them. For a library of documents you query often, a document-chat app is the better tool.
Is Bineret connected to Vane?
No. This page describes Vane and links to the source archive GitHub generates for the project's tag.
Tutorials
Run with Docker (recommended)
docker run -d -p 3000:3000 -v vane-data:/home/vane/data --name vane itzcrazykns1337/vane:latestOpen http://localhost:3000 and the setup screen asks for your model provider, API keys and models. The image includes SearxNG, so nothing else is needed. The volume keeps your settings and uploads between restarts.
Already running SearxNG?
Use the slim image and point it at your instance:
docker run -d -p 3000:3000 -e SEARXNG_API_URL=http://your-searxng:8080 \
-v vane-data:/home/vane/data --name vane itzcrazykns1337/vane:slim-latestYour SearxNG must have JSON output enabled and the Wolfram Alpha engine switched on.
Run exactly 1.12.2 from this archive
unzip Vane-1.12.2.zip
cd Vane-1.12.2
docker build -t vane .
docker run -d -p 3000:3000 -v vane-data:/home/vane/data --name vane vaneWithout Docker, install SearxNG yourself (JSON format and Wolfram Alpha enabled), then:
npm i
npm run build
npm run startConnect a local model through Ollama
In Vane's settings, set the Ollama API URL. From inside Docker on Windows or macOS use http://host.docker.internal:11434. On Linux, make Ollama listen on the network: add Environment="OLLAMA_HOST=0.0.0.0:11434" to /etc/systemd/system/ollama.service, then run systemctl daemon-reload and systemctl restart ollama, and use your machine's address.
If it says no chat model is configured
For a local OpenAI-compatible server, check three things: it listens on 0.0.0.0, not 127.0.0.1; the model name matches what the server has loaded; and the API key field is not empty, even if your server ignores it.
Support
Where support comes from
Vane is developed by ItzCrazyKns and contributors. The README and the installation docs cover setup and updating, bugs go to GitHub issues, and the project runs a Discord community for questions.
What Bineret covers
This page. If the download link is broken or something written here is wrong, tell us and we will fix it. We do not host Vane, run SearxNG, or provide model keys.
No warranty
The MIT licence provides the software as-is. Answers are generated from web pages the model read, so check the cited sources before relying on anything important.
