Best VPS for Open WebUI (2026): Light App, Heavy Backend

Open WebUI keeps growing in popularity as the ChatGPT-replacement UI for self-hosted LLMs. The hosting confusion comes from people conflating Open WebUI’s needs with the inference backend’s needs. Open WebUI is light, the model is what eats the resources.

Here is the realistic VPS shortlist after running Open WebUI in production for six months across personal and team deployments.

The Honest Resource Picture

Open WebUI is two pieces: a SvelteKit frontend that compiles to static assets and a Python (FastAPI) backend that handles user management, chat history, and the proxy to your inference endpoint. Total resident memory at idle: 250 to 400 MB.

For a 10-user team with active chat usage, expect 500 MB to 1 GB resident memory plus the database (SQLite for hobby, Postgres for production). The app scales to many users without significant additional resources because the heavy lifting happens on the inference backend.

The disk story is mostly about chat history. Active users accumulate 10 to 50 MB per month of chat logs. Plan for the volume to be predictable and small unless you have heavy ingestion of long documents.

VPS Comparison for Open WebUI

Hetzner CPX11: For solo and small team use

At 5.18 EUR a month, the CPX11 is enough for Open WebUI when your inference happens elsewhere (external API, separate GPU machine, hosted model). 2 GB RAM holds the app, the SQLite database, and a small user count comfortably.

This is the right pick for personal use or small teams (under 5 users) that point Open WebUI at OpenAI, Anthropic, or a separately hosted Ollama instance.

Pros:

• Cheapest realistic Open WebUI setup
• Plenty for the app process plus user management
• Falkenstein latency is fine for European users

Get Hetzner: Hetzner Cloud.

Hetzner CPX21: For active team use

For teams of 10 to 30 users, the CPX21 with 4 GB RAM gives headroom for the chat history database and concurrent web connections. Still cheap at 5.83 EUR a month.

Pick this when Open WebUI is the daily AI tool for a small team and inference happens on a separate machine.

Contabo VPS S: Best value for team setups with co-located Postgres

4.50 EUR for 4 vCPU and 8 GB RAM is hard to beat when you want Open WebUI plus a co-located Postgres for production user management. The shared CPU is fine because Open WebUI is mostly I/O bound on the inference backend, not CPU-bound on its own.

Pros:

• Best raw spec for the money
• 8 GB RAM fits Open WebUI plus Postgres plus headroom for spikes
• 100 GB NVMe holds chat history for years

Real negative: provisioning takes hours, network egress can be slow at peak times.

Get Contabo: Contabo VPS.

Hetzner CCX23: For Open WebUI + Ollama on one machine

When you want everything on one machine (Open WebUI plus Ollama plus models), the CCX23 with 16 GB RAM hits the sweet spot. A 7B model takes 8 GB, Open WebUI plus Postgres takes 1 to 2 GB, leaving headroom.

For team setups where everything needs to live on one machine for simplicity or for privacy. The downside is that you cannot scale the inference and the UI independently.

What I Would Pick

For solo personal use with external model: Hetzner CPX11. For team use with separately hosted Ollama: Hetzner CPX21 or Contabo VPS S. For everything on one machine: Hetzner CCX23. The right split depends entirely on your inference strategy, Open WebUI itself is almost free to host.

Full VPS context lives at the SelfHostVPS comparison. Open WebUI pairs naturally with Ollama, see that guide for the inference-side recommendations.