Your own private AI chat on a server
Open WebUI is a free, self-hosted chat interface that looks and feels like ChatGPT. Paired with Ollama, it runs open AI models on your own server: your chats stay on a machine you control, and you can open it from your phone or laptop. This guide takes about 30 minutes, only copy and paste.
Disclosure: the Hostinger links on this page are referral links. If you sign up, I may earn a commission at no extra cost to you. Any VPS provider with Ubuntu works with these steps.
Be realistic first
- A normal VPS has no graphics card, so models run on the CPU. Small models (1B to 3B parameters) answer at a usable speed; big models are slow. It will not match the top paid AI services in quality.
- RAM decides what you can run. Ollama's own guidance: at least 8 GB RAM for 7B models, 16 GB for 13B. With 4 GB, stick to small models such as llama3.2:1b or llama3.2:3b.
- Already pay for an AI API (OpenAI-compatible)? Open WebUI can use it too, and then a 2 GB server is enough.
1. Prepare the server
Follow steps 1 to 4 of our self-hosting guide (choose Ubuntu, connect with SSH, install Docker, turn on the firewall). Then come back here.
2. Start Open WebUI with Ollama built in
The official image with the :ollama tag bundles both in one container. This version is bound to 127.0.0.1 so it is reachable only through the HTTPS proxy in step 3:
docker run -d --name open-webui --restart always -p 127.0.0.1:3000:8080 -v ollama:/root/.ollama -v open-webui:/app/backend/data ghcr.io/open-webui/open-webui:ollamaYour chats live in the open-webui volume and the downloaded models in the ollama volume, so they survive updates and reboots.
3. Put HTTPS in front
Point a subdomain (for example ai.example.com) to the server IP with an A record. Install Caddy as in the main guide and set /etc/caddy/Caddyfile to:
ai.example.com {
reverse_proxy 127.0.0.1:3000
}systemctl reload caddy4. Create your account right away
Open https://ai.example.com. The first account created becomes the administrator, so do this immediately after the site goes live. In the admin settings you can keep new sign-ups disabled or require approval.
5. Download a model and chat
In Open WebUI, open the model selector or the admin settings for models and pull a small model by name, for example llama3.2:3b (about 2 GB) or llama3.2:1b for 4 GB servers. When the download ends, pick it and start chatting. You can also upload documents and ask questions about them.
6. Keep it healthy
- Updates:
docker pull ghcr.io/open-webui/open-webui:ollama, then remove and re-run the container with the same command (the volumes keep your data). - Disk: each model takes 1 to 5 GB or more; delete models you do not use.
- Backups: provider snapshots, or back up the two Docker volumes.
More free tools
Our 50 Free Tools Kit (PDF, US$ 4.99) lists 50 free alternatives to paid apps, including AI tools you can run yourself.
Get a VPS on Hostinger (referral link)Open WebUI and Ollama are independent open-source projects; this guide is not affiliated with them. Free Tool Daily: one free tool every day. All tools.