Open-source project
ItzCrazyKns/Vane avatar
ItzCrazyKns/Vane

Vane: A Self-Hosted Answer Engine That Keeps Search History Local

Vane is an AI-powered answering engine.

38,210 stars4,303 forksTypeScriptMIT

At a glance

What is it?
Vane is a privacy-focused AI answering engine that runs on your own hardware, combining SearxNG web search with local or cloud LLMs and cited sources. This review covers its architecture, setup paths, limitations, and who should adopt it.
Who is it for?
Adopt Vane if you want a self-hosted, privacy-preserving answer engine that keeps search history on your own disk and supports local LLMs via Ollama. Skip it if you need a managed cloud service or if you are not comfortable running Docker and SearxNG.
Can I use it commercially?
Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
Is it still maintained?
Yes. The repository last received commits 75 days ago.
What is it written in?
Mainly TypeScript, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 14, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • ItzCrazyKns/Vane38,210 ★ · +1,979
Star history
Repository2026-08-182026-11-14
ItzCrazyKns/Vane36.2k38.2k

1,979 Stars in the last 88 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

What Vane Solves and Who It Is For

Vane addresses a specific tension: AI answer engines like Perplexity are convenient but send your queries to third-party servers. Vane is a privacy-focused alternative that runs entirely on your own hardware. It combines web search results with LLM-generated answers and cited sources. The target user is someone who already runs self-hosted services, values control over their search history, and is willing to manage a Docker container and a search backend. The README emphasizes that every search is saved locally, which is a direct answer to the data-retention concerns of cloud-based assistants. It is not a product for non-technical users; the setup requires Docker or manual Node.js builds, and the troubleshooting section assumes you know what an API URL is.

How Vane Works: SearxNG, LLMs, and Search Modes

The architecture is visible from the README and the repository layout. Vane uses SearxNG as its web search backend, which aggregates multiple search engines without exposing your identity. The bundled Docker image includes both Vane and SearxNG, so a single container handles search and the answering engine. You can also point Vane at an external SearxNG instance using the slim image and the SEARXNG_API_URL environment variable. The answering engine connects to LLM providers: local models through Ollama, or cloud providers like OpenAI, Claude, and Gemini. Three search modes exist: Speed Mode for quick answers, Balanced Mode for everyday searches, and Quality Mode for deep research. Widgets like weather, calculations, and stock prices appear when relevant, which suggests a rule-based or model-triggered UI layer, though the README does not specify the mechanism.

Installation: Docker, Slim, and Source Builds

The README gives three installation paths. The recommended one is the full Docker image: docker run -d -p 3000:3000 -v vane-data:/home/vane/data --name vane itzcrazykns1337/vane:latest. That command starts the container with a bundled SearxNG, so no separate search setup is needed. The slim image is for users who already run SearxNG; it requires setting the SEARXNG_API_URL environment variable, and the README warns that your SearxNG must have JSON format enabled and Wolfram Alpha search engine enabled. The non-Docker path involves cloning the repo, running npm i, npm run build, and npm run start, but you must install and configure SearxNG yourself. The setup screen at localhost:3000 is where you enter API keys, model choices, and the SearxNG URL. The troubleshooting section reveals a common failure: local OpenAI-compliant servers must bind to 0.0.0.0, not 127.0.0.1, or Vane will not connect.

Privacy and Data Locality: The Core Trade-off

The privacy claim is the project's main selling point, but it is not absolute. Search history is stored locally, which is a concrete privacy benefit. However, if you use a cloud LLM provider like OpenAI or Claude, your queries and the retrieved search results are sent to that provider. The README does not hide this; it says Vane supports cloud providers, which implies your prompts leave your machine. For true privacy, you would pair Vane with Ollama, but the README has a troubleshooting entry for Ollama connection errors, whose fixes are to check the Ollama API URL and set it for your OS, for example http://host.docker.internal:11434 on Windows and Mac. So the privacy guarantee is conditional: local search history, but not necessarily local LLM inference. This is a trade-off the README does not explicitly spell out, but it is evident from the feature list. Users who need absolute data locality must choose Ollama and accept the configuration friction.

Search Modes and Source Selection: Practical Limits

Vane offers three search modes and source selection (web, discussions, academic papers). The modes are a useful differentiator: Speed Mode for quick answers, Balanced for everyday, Quality for deep research. The README does not define what each mode changes under the hood, so the user must experiment. Source selection is also vague; the README says you can pick web, discussions, or academic papers, but it does not list which search engines SearxNG queries for each category. The widget system is another feature with thin documentation: weather, calculations, stock prices are mentioned, but there is no description of how Vane decides to show a widget or whether it uses a separate API. These gaps mean the practical behavior is not fully predictable from the README. For an engineer evaluating adoption, this is a real limitation: you may need to read the source code or test the UI to understand the exact behavior.

Limitations and Failure Modes

The README's troubleshooting section is the best source of known failure modes. The most prominent is the local LLM connection issue: Vane may report that no chat model providers are configured even when you have one running. The fixes are specific: bind the server to 0.0.0.0, specify the correct model name, and do not leave the API key field empty, even if your server does not require one. Another failure mode is the SearxNG requirement: if you use the slim image, you must ensure JSON format and Wolfram Alpha are enabled, otherwise search will fail. The non-Docker installation is more fragile because you manage SearxNG separately. The README also implies that the bundled Docker image is the only way to get a working SearxNG without extra steps, so users who avoid Docker have a steeper path. These are not theoretical; they are documented in the project's own troubleshooting guide.

Alternatives and the Difference in Approach

The most obvious alternative is Perplexity, which is a hosted answer engine with cited sources. The difference is fundamental: Perplexity is a cloud service, so your queries and search history go to a third party. Vane is self-hosted, so you control where the data lives. Another alternative is running SearxNG alone and combining it with a separate LLM tool, but that requires you to build the integration yourself. Vane bundles the integration, which is its value. There is also the option of using a local LLM with a plain chat interface, but that lacks web search and citations. Vane's approach is to glue SearxNG and LLMs together with a UI that handles search modes, widgets, and history. The trade-off is that you must maintain the infrastructure, whereas Perplexity handles everything. The README mentions that Tavily and Exa support are coming, which would add commercial search APIs as alternatives to SearxNG, but those are not available yet.

Maintenance, Upgrade, and License

The project is MIT-licensed, which means you can modify and redistribute it with few restrictions, and there is no copyleft obligation. The repository shows recent releases, with v1.12.2 from April 2026 as the latest release and the last push on 2026-09-01. The README points to an installation documentation folder for updating, but it does not describe the upgrade process in detail. The Docker volume flag -v vane-data:/home/vane/data persists your data across container updates, which is a practical hint that upgrades are meant to be done by pulling a new image and recreating the container. The maintenance cost is moderate: you need to keep the Docker image updated, manage the persistent volume, and ensure your SearxNG instance (if external) stays compatible. The bundled image simplifies maintenance because it includes SearxNG, but it also means a larger image and a single point of failure. The license is permissive, so forking is possible, but the project's roadmap is community-driven through Discord, which is a risk if you depend on long-term support.

Editorial conclusion

Adopt Vane if you want a self-hosted, privacy-preserving answer engine that keeps search history on your own disk and supports local LLMs via Ollama. Skip it if you need a managed cloud service or if you are not comfortable running Docker and SearxNG. Before deploying, verify that your SearxNG instance has JSON format and Wolfram Alpha enabled, and confirm that a local OpenAI-compatible server binds to 0.0.0.0. The project is MIT-licensed, with v1.12.2 from April 2026 as its latest release, but the README itself warns about connection pitfalls, so test the setup screen first.

Frequently asked questions

What is an open-source AI search engine like Vane?

Vane is one. It carries the MIT license, runs entirely on your own hardware, and combines web knowledge through SearxNG with answers from Ollama or a cloud provider such as OpenAI, Claude, Gemini, or Groq, returning cited sources. You start it with a single docker run command and configure it in a setup screen at http://localhost:3000.

What is perplexica, and what does Vane say about it?

Vane describes itself as a privacy-focused answering engine on Next.js that runs on your own hardware with Ollama and cloud provider support.

Is there an open-source version of Perplexity, and does Vane fit?

Vane is open source under the MIT license and self-hosted. What it does specify is that it runs on your hardware, bundles SearxNG for search, and cites the sources behind each answer.

Is perplexica free, and what license does Vane carry?

The package manifest and the repository both carry the MIT license, and the sponsor section says contributions help keep the project free and open source, with Exa listed as a web search, crawling, deep research, and answer API provider.

Official sources

  1. Official README
  2. Project repository
  3. Release notes
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/itzcrazykns-vane.svg)](https://hysenlabs.com/projects/itzcrazykns-vane)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/itzcrazykns-vane.svg)](https://hysenlabs.com/projects/itzcrazykns-vane)

Was this useful?