Who Vane is for#
Privacy-conscious researchers handling sensitive topics
Vane keeps every query, citation, and search history local. For researchers working on competitive intelligence, legal questions, or personal health topics, avoiding a third-party platform's log trail is worth a one-time Docker setup. All results cite their sources, so the research trail stays on your own machine.
Skip if:
If you need to share a search session or collaborate in real time with teammates, Vane currently has no multi-user or sharing features. A managed tool fits better for collaborative research workflows.
Developers integrating AI search into internal tools
Vane exposes an API for running searches and getting cited answers programmatically, with support for multiple models. Teams building internal knowledge tools, documentation assistants, or research pipelines can query Vane's API instead of paying per-call to a managed AI search service.
Skip if:
If your use case requires enterprise-grade SLAs, managed uptime, or a vendor-hosted API, Vane's self-hosted requirement means you own the reliability work. A managed provider is the simpler path when infrastructure is not your team's strength.
Homelab operators who want AI search without a subscription
One Docker command starts Vane with SearxNG bundled, accessible at localhost:3000. It can be added to any browser as a default search engine via a custom search URL, making it a drop-in replacement for Google or Bing for day-to-day browsing on a local network.
Skip if:
If you do not have a server or NAS running Docker, the self-hosting requirement is a real barrier. Perplexity AI's free tier is a simpler starting point until you have infrastructure ready.
Teams reviewing confidential documents without cloud uploads
File upload support lets users ask questions about PDFs, text files, and images without sending documents to a third-party cloud service. For legal, financial, or compliance teams where document confidentiality matters, running Vane on an internal server keeps sensitive files off vendor infrastructure entirely.
Skip if:
Vane has no multi-user access controls yet. A team larger than a few people sharing document access needs an authentication layer added externally before this use case is secure at a team scale.
The problem it solves#
Commercial AI answering engines are convenient, but every query you send passes through a third-party platform. Your research patterns are logged, your questions become part of a vendor's data pipeline, and you have no control over retention or what the data is used for. For sensitive research topics, competitive intelligence, or simply not wanting a company to build a profile from your search habits, there is no privacy setting that fixes this: the data leaves your device by design.
Getting cited AI search answers locally requires either a managed service that logs everything, or piecing together a local LLM with a separate search integration and hoping the integration stays maintained. Most self-hosters who try the DIY route hit configuration friction fast: search APIs need keys, models need hardware, and getting the pieces to talk to each other reliably is its own engineering project.
How it solves it#
Multi-provider AI model support
Connect local LLMs through Ollama or cloud providers including OpenAI, Anthropic Claude, Google Gemini, and Groq. Models are swappable per session, so you can run sensitive queries through a local Ollama model and route less sensitive searches through a cloud API when you need more capability. No single vendor lock-in.
Three search depth modes
Speed Mode returns quick answers for simple lookups; Balanced Mode handles everyday research; Quality Mode runs a deeper multi-source pass for complex topics. The mode is selectable per query, letting you trade response time for answer depth depending on what the search demands.
Privacy-preserving web search via SearxNG
All web queries route through a bundled SearxNG instance, which aggregates results from multiple search engines without identifying you to any of them. The Docker image ships SearxNG alongside Vane, so no external search API key is needed to get started. Users with an existing SearxNG instance can point Vane at it using the slim image.
File upload and document Q&A
Upload PDFs, text files, or images and ask questions about their contents. Vane answers from the uploaded file rather than a web search, making it practical for private document research or reviewing a contract without sending files to a cloud service.
Domain-specific search
Restrict a query to a single site or domain when you know exactly where to look, such as a specific documentation site or a research archive. Results and AI answers draw only from the specified domain for that query, cutting noise from unrelated sources.
Strengths and trade-offs#
Strengths
- Full data ownership with MIT licenseVane is MIT licensed, meaning you can run it on your own infrastructure, modify the source, and use it commercially without restriction. Search history is stored locally in a Docker volume and never leaves your server. Unlike Perplexity AI, there is no telemetry to a vendor and no data retention policy to audit.
- Bundled SearxNG requires no search API keyThe default Docker image bundles SearxNG alongside Vane, so multi-engine web search works out of the box without signing up for any external search API. Tavily and Exa are listed as upcoming search source options for users who want API-backed retrieval alongside the bundled path.
- Complete local inference with OllamaOllama support lets you run the entire stack on a local machine with no outbound API calls at all. A home server or workstation with a capable GPU can run Vane plus a local model, giving you cited AI search that never contacts a cloud provider for inference.
Trade-offs
- -No built-in authenticationVane has no login system as of the current release; adding authentication is on the project roadmap. Running the service on a network-accessible port without a reverse proxy with auth leaves the interface open to anyone who can reach it. Self-hosters need to add their own auth layer before exposing Vane outside a private local network.
- -SearxNG configuration required for advanced setupsUsing your own SearxNG instance instead of the bundled one requires JSON format and Wolfram Alpha enabled in SearxNG settings. On Linux, Ollama connections need the host's private IP rather than localhost due to Docker networking, which catches users new to container setups. The default bundled image avoids most of this, but deviating from it adds configuration steps.
- -356 open issues on an actively growing codebaseThe GitHub tracker shows 356 open issues, which is expected for a project with 37,000+ stars but signals that edge cases accumulate faster than they close. The project is actively maintained, with the last push in September 2026, but users should expect occasional rough edges as new features roll out.
Vane vs alternatives#
Vane vs Perplexity AI
Both tools answer questions by searching the web and citing sources. The deciding factor is who controls the infrastructure and what happens to your queries.
| Feature | Vane | Perplexity AI |
|---|---|---|
| License | MIT | Proprietary |
| Self-hosting | Yes | No |
| Local LLM support | Yes (Ollama) | No |
| Search privacy | Queries through self-hosted SearxNG | Queries sent to Perplexity's servers |
| Cost | Server costs only | Free tier with paid plans |
| Authentication | Not built-in (roadmap) | Managed account |
Vane is the better choice when data privacy is the primary concern. Queries never leave your server: SearxNG aggregates results without revealing your identity to underlying search engines, and model inference runs locally if you use Ollama. For individuals and teams who handle sensitive research or simply prefer not to feed their search habits into a commercial platform, the self-hosted setup pays off after one Docker command.
Perplexity AI is still the better fit when you want a managed, no-infrastructure experience. There is nothing to install or maintain; it works from any browser on any device. For users without a server running Docker, Perplexity AI's free tier is a more practical starting point than self-hosting Vane. The trade-off is that your queries and research history live on Perplexity's servers under their terms of service.
Quick start#
Self-hosting uses Docker with SearxNG bundled in the same image.
```bash
docker run -d -p 3000:3000 -v vane-data:/home/vane/data --name vane itzcrazykns1337/vane:latest
```What it's built on#
- Languages
- TypeScript
- Frameworks
- Next.jsReact
- Search
- SearXNG
FAQ#
Can Vane run entirely offline without any cloud API calls?
Yes, when you pair Vane with a local LLM through Ollama, all model inference happens on your machine with no outbound API calls. Web search through SearxNG still reaches public search engines unless you configure SearxNG to use only internal or intranet indexes. For a fully air-gapped deployment, you need both Ollama and a SearxNG instance pointed at local sources.
Does Vane require a GPU to run?
Vane itself runs on any server with Docker and has no GPU requirement. GPU usage depends on how you configure the AI backend: if you use Ollama with a local model, a GPU speeds up inference significantly but is not strictly required for smaller models. If you connect to a cloud API like OpenAI or Groq, the GPU requirement moves entirely to those providers.
How does Vane compare to Perplexity AI for everyday search?
Both tools return cited web answers, but Perplexity AI is a managed cloud service while Vane runs on your own hardware. Vane adds local LLM support and keeps all data on your server under MIT license; Perplexity AI requires no infrastructure and works from any browser with no setup. Vane is the better choice when data privacy or cost at scale matters more than convenience.
How do I update Vane to the latest version?
Pull the latest Docker image and restart the container. Persistent data is stored in the vane-data volume mounted with the -v flag, so your configuration and search history survive the update. The README links to the installation documentation for full update steps.
What search engines does Vane use to find results?
Vane routes all searches through SearxNG, a meta-search engine that aggregates results from multiple underlying engines without revealing your identity to any of them. The default Docker image bundles SearxNG in the same container, so no separate setup is required. Tavily and Exa are noted in the README as upcoming additional source options.
Similar open-source tools#
hyperresearch
Deep research agent with a persistent vault for Claude Code
Local Deep Research
Your AI research assistant, fully local and encrypted.
Scira
Open source AI search engine that retrieves cited sources
Cloudflare Os
Open source AI workspace with sandboxed apps and Gatekeeper security
Firebase Ios Sdk
Open source Apple SDK for Firebase auth, data, push and crashes
openGym
Self-hosted gym tracker with no subscription or telemetry