Vane

Want an AI search interface that you can run yourself? Vane is an open source answering engine with cited answers, local-model options, and configurable cloud providers. This guide helps you decide whether operating your own instance is worth the control it gives you.

View the upstream project · Start locally · Compare your options

Independent guide. Documentation checked October 23, 2026. We have not benchmarked or deployed Vane for this page.

What does Vane do?

Vane's repository documents source-backed search, Ollama and cloud-model connections, and a Docker image with SearxNG included. The code is MIT licensed. Start with the project README for supported options and its license.

That makes it a candidate for someone who wants to operate an answering interface, choose a model, and inspect the application. It does not remove the need to evaluate retrieved sources or maintain the service.

Vane vs Perplexity: choose by the job you want to own

Perplexity also provides conversational answers with linked sources. The useful distinction here is your operating responsibility, not an unsupported claim that one gives better answers.

This is a decision framework, not a feature-equivalence or performance test. Review each product's current documentation before choosing.

Start a local Vane instance with Docker

Install and start Docker first. The command below follows the upstream image and data-volume setup, with the browser port deliberately bound to your own machine:

docker run -d \
  --name vane \
  -p 127.0.0.1:3000:3000 \
  -v vane-data:/home/vane/data \
  itzcrazykns1337/vane:latest

Open http://localhost:3000 on that machine and complete the setup screen. Choose your available model provider and supply credentials only to your own instance. For deployment choices and updates, use the installation documentation.

The command downloads and runs software, creates a container, and persists data in a Docker volume. It is documentation, not a command executed by this guide. Before a longer-lived deployment, record the image version or digest you actually use instead of relying on an unchanged meaning of latest.

Check that the first search actually works

1. Confirm the interface loads and a model is selected.
2. Ask a narrow question with an answer you can independently check.
3. Open the cited pages. Check whether each supports the sentence attached to it.
4. Change the date or source restriction in your question and look for stale results.
5. Record the model, settings, runtime, and any provider charges for the task.

Try this original evaluation prompt:

Find the current installation requirements for a library I specify.
Prefer its official documentation. Link the pages you used, identify the
documented version, and separate confirmed requirements from assumptions.
If the sources disagree, show the disagreement rather than guessing.

Self-hosting and privacy: trace the complete request

Running an interface on your machine does not by itself prove that every part of a search stays there. Consider the model endpoint, search backend, fetched websites, logs, and any uploaded files separately.

Before using sensitive information, draw the path from query to search service to model to stored history. Confirm which steps are local, what is sent externally, and what retention applies. Local models can change that path, but web retrieval still involves network services.

Treat public deployment as a separate project: decide authentication, network exposure, updates, backups, and provider limits before letting other people submit queries. The localhost command above is a starting point for one operator, not a public hosting recipe.

What does it cost?

The repository's software license is not a statement that operating it has no cost. Budget for hardware or a server, model usage if applicable, storage, and time spent maintaining the instance. Test a representative batch of searches and calculate cost per acceptable answer, including retries.

Troubleshoot before changing several settings

Capture the smallest failing query and its settings before opening an upstream issue. Remove keys and private content from shared diagnostics.

Frequently asked questions

Is Vane a drop-in copy of Perplexity?

No equivalence has been tested here. Evaluate the workflow you need, including source quality, model behavior, operating effort, and cost.

Does this page host Vane or a free search demo?

This page is a setup and decision guide. The application runs in your own environment; no query is submitted from this page.

Can I use a local model?

The upstream documentation lists Ollama support. Check the configuration and hardware requirements of the particular model you choose rather than assuming every local model will fit your machine.

Should I choose Vane?

Choose it for a small evaluation when control and self-hosting are concrete requirements. If you mainly want immediate answers and do not want to operate software, compare hosted options first.

Review the upstream setup guide