Vane (formerly Perplexica)

What is a vane?

Vane is the current name for the former Perplexica project, maintained by Kushagra Srivastava (GitHub username ItzCrazyKns) and community contributors. The old repository has been redirected to Vane, so both names refer to the same project. It integrates web search, content retrieval, and large-model responses into a self-deployable search interface, making it suitable for technical users who want to control their choice of models and deployment methods.

As of October 26, 2026, the official GitHub repository has accumulated approximately 37,800 stars and is licensed under the MIT License. The high number of stars reflects community interest but does not guarantee that every answer is accurate, nor does it mean the project was just released today. The latest official release is v1.12.2, dated April 10, 2026.

Key Features and Benefits

• Search Q&A from the source:Include citations when generating answers so that readers can go back to the original source to verify the information, rather than just receiving a summary that cannot be traced back to its source.
• A variety of models to choose from:It can integrate with Ollama's on-premises models and also supports multiple cloud-based model providers, making it easy to configure based on hardware, cost, and capability requirements.
• Search Mode:Offers options such as "Speed," "Balanced," and "Quality," allowing you to balance response speed with more in-depth processing.
• Source and Domain Control:Supports searches across web pages, forums, and academic sources, and allows you to specify domains to narrow your research scope.
• Additional Capabilities Beyond Search:Includes image and video search, Q&A with file uploads, search history, and some informational cards. Actual results depend on the selected model, search service, and deployment environment.

Applicable Scenarios

Vane is ideal for developers building personal search assistants, technical teams validating local models through Q&A, and researchers conducting preliminary analysis of viewpoints across multiple web pages. It is closer to a ready-to-use, interactive product than a simple web crawler or parsing library, but it still requires deployment, model configuration, and service maintenance. If you’re simply looking for a low-maintenance, everyday search tool, a hosted service is usually more convenient.

When it comes to research, the most useful approach is not to “draw conclusions for you,” but to integrate searching, reading, and citing into a single process. Important facts should still be verified by clicking through to the source, especially dates, numbers, policies, and controversial conclusions.

How to Deploy and Get Started

1. Set up Docker according to the official installation documentation, choosing either the full image or the minimal installation option suitable for your existing SearxNG environment.
2. Use the official Vane image to start a local service and persist data; the default example is accessible at localhost:3000.
3. On the settings page, select a model provider and configure credentials for a local Ollama instance or your own cloud model; do not post your keys to a public repository.
4. First, compare the different search modes using questions that do not contain sensitive information to verify that the citations correspond to the answers.
5. Before deploying to the public, ensure that access protection, updates, and data management are in place. Do not expose locally running examples directly as unprotected public services.

How do I choose among related projects?

Vane is closer to a final Q&A interface, while the other two are geared toward data input and processing. It makes more sense to base your choice on whether you need a ready-made interface, the data type, and maintainability, rather than just looking at the number of stars.

Costs, Privacy, and Limits of Capability

Open-source at MIT does not mean zero operating costs: on-premises models require hardware resources, and cloud-based models may charge per call. Self-hosting does not equate to being completely offline or ensuring absolute privacy; online searches involve interactions with search engines and external websites, and when using cloud-based models, relevant content is also sent to the model provider. Before using sensitive data, you should clearly understand the complete data flow and the provider’s policies.

Please note that the accuracy of the responses is not guaranteed. Results may be affected by the quality of upstream search data, model capabilities, context length, and network connectivity. The Tavily and Exa integrations marked as “coming soon” in the official roadmap should not be considered supported features. Release notes should be reviewed in conjunction with the actual installed version.

Official Portal

GitHub Repository and Installation Documentation - v1.12.2 Release Notes - MIT License

Screenshots and Logos: Vane Project, Copyright © 2026 ItzCrazyKns, MIT License. Links to the official repository and license are retained; the answers shown in the interface examples do not serve as factual basis for this article.