Deploy and Host Vane on Railway

Vane — formerly Perplexica, renamed in March 2026 — is the most popular open-source self-hosted alternative to Perplexity AI, with 20k+ GitHub stars. Ask a question and it searches the live web through its own bundled private SearXNG, reranks the results, and streams a cited answer with source images and videos. Connect Claude, GPT-4o, Gemini, Groq, Ollama, or LM Studio — you control the model, the search index, and the data.

One Docker image. No telemetry. No rate limits. No $20/month Perplexity Pro subscription. Self-host on Railway for ~$10/month flat with persistent chat history and provider config.

What This Template Deploys

Service	Purpose
Vane (single container)	The complete stack — Next.js frontend, answer engine, and a private bundled SearXNG metasearch engine, served on port `3000`
Persistent Volume (4 GB)	Mounted at `/home/vane/data` — chat history, uploaded files, and provider configuration survive every redeploy

SearXNG runs inside the same container on internal localhost:8080 — nothing extra to wire up. No PostgreSQL, no Redis, no gateway service. Bring one LLM API key and start searching.

About Hosting Vane

Perplexity Pro costs $20/month, logs every query, and locks you to their models. Vane gives you the same experience — live web search, cited answers, focus modes — on your own infrastructure with full query privacy. Since the March 2026 rename from Perplexica, the project has kept shipping: a redesigned setup wizard, deep research mode, file uploads, domain-limited search, and a custom streaming engine that replaced LangChain entirely.

Railway runs the bundled image with automatic HTTPS and a persistent volume. The image ships Playwright and Chromium for headless page rendering, so allocate at least 4 GB RAM to avoid out-of-memory failures during cold starts — this template pre-configures that.

Typical cost: ~$10/month on Railway's Hobby plan. Perplexity Pro is $20/month per user. Groq and Gemini both offer generous free API tiers, so your per-query model cost can be zero.

Deploy in Under 5 Minutes

Click Deploy on Railway — the Vane container builds automatically (~3–4 minutes; the image is large because it bundles SearXNG and Chromium)
Open your Railway-assigned URL — the Setup Wizard loads on first visit
Pick a provider (OpenAI, Anthropic, Gemini, Groq, Ollama, or LM Studio) and paste your key
Choose a chat model and an embedding model, confirm, and ask your first question
Settings persist to the volume — redeploys never repeat the wizard

No SearXNG wiring. No config files. No database setup.

Common Use Cases

Self-hosted alternative to Perplexity AI — live web search with cited answers at ~$10/month flat instead of $20/month per user, with zero query logging
Perplexica users migrating to the new Vane image — same project, new name since March 2026; this template runs the current vane:latest image with the bundled SearXNG
Private research over the live web — queries route through your own SearXNG instance; no search history sold, no profile built, no telemetry leaving the container
Academic and focused research — focus modes constrain search to academic papers, YouTube, Reddit, Wolfram Alpha, or specific domains you name
Ask questions about your own documents — upload PDFs, text files, and images alongside web search; Vane answers from both
Team search on a shared instance — one deployment, one model bill, search history saved locally for the whole team

Configuration

Variable	Required	Description
—	—	No environment variables required — all provider setup happens in the browser Setup Wizard
`SEARXNG_API_URL`	Optional	Only for the `slim` image variant if you run SearXNG separately — the default image bundles it
`PORT`	Auto-set	Railway injects the port; Vane serves on `3000`

All configuration — API keys, chat model, embedding model — is entered in the Setup Wizard and persisted to /home/vane/data/config.json on the volume. Keys never live in environment variables or version control.

Vane vs. AI Search Alternatives

	Vane (Railway)	Perplexity Pro	Morphic	Open WebUI + search
Monthly cost	~$10 flat	$20/user	Hosted/paid	~$5 (self-hosted)
Query privacy	✅ Your SearXNG	❌ Logged	❌ Cloud	✅ Configurable
Model choice	✅ Any provider + Ollama	❌ Their models	⚠️ Limited	✅ Any
Cited answers	✅ Inline + sources	✅ Yes	✅ Yes	⚠️ Basic
Focus modes	✅ Academic, YouTube, Reddit, Wolfram	✅ Some	❌ No	❌ No
File uploads	✅ PDF, text, images	✅ Yes	❌ No	✅ Yes
Search bundled in image	✅ SearXNG built in	N/A	N/A	❌ Separate setup
Open source	✅ MIT	❌ No	✅ Partial	✅ Yes

Dependencies for Vane Hosting

Railway account — Hobby plan; allocate 4 GB RAM (~$10/month typical usage)
One LLM API key — Groq and Gemini free tiers work; Ollama for fully local inference
Nothing else — SearXNG and storage are bundled

Deployment Dependencies

Vane GitHub Repository — source, formerly Perplexica
Vane Docker Image — bundled image
Groq Console — free-tier API keys
Railway Volumes Documentation — persistent storage

Implementation Details

This template deploys itzcrazykns1337/vane:latest — the post-rename image that bundles the Next.js app, the answer engine, and a private SearXNG configured with JSON output and the Wolfram Alpha engine enabled. A 4 GB volume at /home/vane/data persists chat history, uploads, and config.json.

The image is ~3.6 GB because it ships Playwright and Chromium for headless web rendering — this powers deep research mode and page scraping. The template allocates 4 GB RAM to prevent OOM during cold starts. Vane's v1.12 line removed LangChain for a custom streaming and function-calling implementation, so provider behaviour is handled natively per connection.

Frequently Asked Questions

Is Vane the same as Perplexica? Yes. Perplexica was renamed to Vane in March 2026 — same maintainer, same codebase, continued development. The Docker image moved from itzcrazykns1337/perplexica to itzcrazykns1337/vane and internal paths changed from /home/perplexica to /home/vane. The old GitHub URL redirects to the Vane repository.

Do I need an OpenAI API key? No — any one provider works: Anthropic Claude, Google Gemini, Groq, OpenAI, Ollama, or LM Studio. Groq and Gemini both have generous free tiers, so you can run Vane with zero per-query model cost. For fully local inference, point the wizard at a remote Ollama endpoint.

Why does this template need 4 GB RAM? The bundled image ships Playwright and Chromium for headless page rendering, which Vane's deep research and scraping features use. Under 4 GB, cold starts can hit out-of-memory failures. This template pre-allocates the right resources so deploys succeed on the first try.

Is my search history private? Yes. Queries go through the bundled private SearXNG, answers are generated by the model you chose, and history is stored on your Railway volume. No telemetry leaves the container. Note that upstream search engines still receive anonymised queries from SearXNG, and your LLM provider sees the prompts you send it.

Do I lose chat history and settings on redeploy? No. History, uploads, and provider configuration persist to /home/vane/data on the volume. Redeploys and version updates skip the Setup Wizard and keep everything.

How do I update Vane? Trigger a redeploy to pull vane:latest, or pin a specific tag in the service settings. Your volume data is unaffected. Watch the Vane GitHub releases for breaking changes — the project ships actively.

Why Deploy and Host Vane on Railway?

Railway is a singular platform to deploy your infrastructure stack. Railway will host your infrastructure so you don't have to deal with configuration, while allowing you to vertically and horizontally scale it.

By deploying Vane on Railway, you get the most popular open-source Perplexity alternative — live web search, cited answers, focus modes, and file uploads — in one container with bundled private search, at ~$10/month flat with no query logging and no per-user subscription.