Independent guide to self-hosted AI search
Vane: an open-source Perplexity alternative
Vane is an open-source answer engine you run on your own hardware. It uses SearXNG for web search and a model you choose to produce answers with sources. It suits people who want control over their search setup and are willing to maintain it.
This is an independent guide, not the Vane project or an official download site. Visit the official repository.
Own the setup, accept the setup work.
Vane can run with a local model through Ollama or connect to a cloud model provider. Its Docker image bundles SearXNG. You configure the model, keep the service running, and check the answer sources yourself.
Decide before you install
Where Vane fits
Consider Vane if you want
- A self-hosted interface for answer-style web search.
- Choice between a local Ollama model and supported cloud providers.
- Control over the search and model configuration.
Plan for these trade-offs
- You need to install and maintain the service.
- Answer quality depends on the model and retrieved results.
- Using a cloud model sends prompts to that provider; web search still reaches upstream search services.
Start with your question
Choose, understand, or install
Six guides answer the six questions examined in our October 2026 ChatGPT research. Each page links to primary product sources; product testing is labeled separately.
Quick answers
Questions people ask before switching
Is Vane the same project as Perplexica?
Yes. The project's README says Vane was formerly known as Perplexica. Older tutorials may use the former name. Check commands against the current repository.
Can Vane work with Ollama?
Yes. Vane lists Ollama among its supported local model options. Its published setup notes include a Docker host address for Windows and macOS. See the Ollama connection guide.
Do I need a paid search API?
The official all-in-one Docker image bundles SearXNG, so its default setup does not require a separate paid search API. A chosen cloud model may have its own charges.
Will a local model keep every search on my computer?
No. A local model can keep model inference on your machine, but web retrieval can still contact external search services. Review your complete setup before making privacy claims.
Source & editorial method
Check the primary source
Our October 21 research used six separate ChatGPT conversations about alternatives, self-hosting, Vane's identity, comparison, and Docker setup to identify the questions worth answering. Product facts and commands on this site are checked against the linked project documentation as of October 22, 2026. The guide has not yet published a hands-on performance benchmark.