What is Vane (formerly Perplexica)?
Vane is the project called Perplexica until March 2026: an open-source take on Perplexity that you run yourself. You ask a question, it searches through a bundled SearXNG metasearch instance, reads the results and writes an answer with citations. Models come from Ollama and other local servers, or from OpenAI, Anthropic, Google Gemini, Groq and other providers.
It offers Speed, Balanced and Quality modes, search over the web, discussions or academic papers, image and video search, file uploads, search limited to chosen domains, widgets for weather and stock prices, a Discover feed and saved search history. An HTTP API (POST /api/search) exposes the same pipeline to other programs.
The code is MIT-licensed TypeScript on Next.js. One Docker command starts Vane and SearXNG together on port 3000.
Is Vane (formerly Perplexica) worth it?
Vane does the Perplexity routine on your own hardware: search, read, answer with sources. It has 38,242 GitHub stars, nearly all earned under the Perplexica name; the rename landed in a commit on 9 March 2026, and the Docker image is now itzcrazykns1337/vane. Version 1.12.2, released on 10 April 2026, rebuilt the Quality mode around a repeated reason, search, scrape and extract loop, added a Chromium-based scraper and uses embeddings to trim search results so they fit the model's context window.
Privacy depends on the model you choose. Searches go out through SearXNG, which queries other engines without your accounts; the question and the scraped page text then go to whichever model you configure. With Ollama or another local server, only the search queries leave your network. With OpenAI or Anthropic, your questions and the page text reach them under their terms.
There is no authentication. "Adding authentication" sits in the README's list of upcoming features, while the same README says the app works across your network and through port forwarding. Anyone who can reach port 3000 can run searches on your API keys, read the saved history and open the settings screen. Keep it on localhost, behind a VPN, or behind a reverse proxy that adds a login. The troubleshooting section also tells Linux users to make Ollama listen on 0.0.0.0, which puts Ollama's own unauthenticated API on the local network unless a firewall blocks it.
Momentum is the other concern. Of 47 commits in 2026, the last code change was on 11 April; the only commit since is a README edit on 1 September, and 384 issues and pull requests are open. Tavily and Exa search support is still listed as coming soon. Next to GPT Researcher, which writes long cited reports, Vane is the quicker everyday search box; next to Open WebUI with web search switched on, it is a more focused search page but not a multi-user app.
Where it earns its place
A privacy-minded developer with a 24GB GPU running Ollama
Replace Perplexity for daily technical questions without sending queries to an AI company
Vane runs on the same machine, uses a local model for answers and the bundled SearXNG for search, and becomes the browser's default search engine through http://localhost:3000/?q=%s. Cost is electricity; queries reach only the search engines SearXNG calls.
A two-person research consultancy
Answer client questions with cited sources from academic papers and chosen sites
Academic search and domain-limited search narrow the sources, Quality mode runs the deeper scrape-and-extract loop, and answers keep their citations for the report. Behind a VPN, with a cloud model billed per token by the provider.
An engineer building an internal tool that needs web answers
Call a search-and-answer pipeline from a script instead of a paid search API
POST /api/search with a query, model and source choice returns an answer and its sources. GET /api/providers lists configured models. It needs a network-level access control because the API has no keys of its own.
Before you commit
- Setup
- docker run -d -p 3000:3000 -v vane-data:/home/vane/data itzcrazykns1337/vane:latest, then add API keys and models in the setup screen. A slim image works with an existing SearXNG that has JSON output and Wolfram Alpha enabled.
- Integrations
- Ollama, Lemonade and OpenAI-compatible local servers; OpenAI, Anthropic, Google Gemini and Groq; SearXNG for search; an HTTP API with /api/chat, /api/search and /api/providers.
- Data and privacy
- Chats, messages, uploads and API keys are stored in the Vane data volume on your machine. Questions and scraped text go to the model provider you configure. There is no login.
- Cost at scale
- No licence or hosted fee. Cost is the server plus model usage: nothing extra with local models, per-token billing with cloud providers, and Quality mode makes more model calls per question.
- Limits
- No user accounts or authentication; last code change 11 April 2026; Tavily and Exa search not yet supported.
What’s good
- One Docker command runs the app and SearXNG together
- Works fully with local models through Ollama, so only search queries leave the network
- Speed, Balanced and Quality modes, plus file uploads and domain-limited search
- Search API usable from scripts and other apps
What’s not
- No authentication; anyone who reaches port 3000 can use your keys and history
- No code changes since 11 April 2026, with 384 open issues and pull requests
- Ollama setup on Linux asks you to expose Ollama on all interfaces
- Answer quality depends heavily on the model and on SearXNG's results
Best for
Individuals and homelab owners who want a private Perplexity-style search page backed by local or their own API-keyed models, on a machine only they can reach.
Vane (formerly Perplexica) alternatives
Vane (formerly Perplexica): common questions
- Is Vane (formerly Perplexica) worth it?
- A capable private Perplexity clone for one person on a trusted network, held back by having no login at all and code that has not changed since April. We rate it 3.6 out of 5.
- How much does Vane (formerly Perplexica) cost?
- Free and open source under MIT. No hosted or paid tier; you pay for any cloud model API you connect.
- What are the downsides of Vane (formerly Perplexica)?
- No authentication; anyone who reaches port 3000 can use your keys and history; no code changes since 11 April 2026, with 384 open issues and pull requests; ollama setup on Linux asks you to expose Ollama on all interfaces; answer quality depends heavily on the model and on SearXNG's results.
- Who is Vane (formerly Perplexica) best for?
- Individuals and homelab owners who want a private Perplexity-style search page backed by local or their own API-keyed models, on a machine only they can reach.
- How much does Vane (formerly Perplexica) cost for a team or in production?
- No licence or hosted fee. Cost is the server plus model usage: nothing extra with local models, per-token billing with cloud providers, and Quality mode makes more model calls per question.
- Is Vane (formerly Perplexica) safe for company data?
- Chats, messages, uploads and API keys are stored in the Vane data volume on your machine. Questions and scraped text go to the model provider you configure. There is no login.
- What are the best Vane (formerly Perplexica) alternatives?
- The closest alternatives we have reviewed are GPT Researcher, Open WebUI, Khoj and RustXNG.
Answers checked Nov 18, 2026.