AI
Self-Host Private AI Chat On Your Own Server
Run DeepSeek And Other Open Models On Your Own Server
Open WebUI with Ollama beside it, so AI models run on your own server instead of a provider's: no API key, no per-message bill, and conversations never leave the box. On an ordinary server only small models are practical — DeepSeek-R1 1.5B answers at a usable speed; 7B and 8B work but slowly.
What ZipLoom Does When You Launch It
The Publisher's Own Image
ZipLoom runs Open WebUI Inc. (chat) and Ollama (model runtime)'s official image (ghcr.io/open-webui/open-webui:0.12) on a server you rent from Vultr, DigitalOcean, Contabo or Amazon Lightsail. Nothing is rebuilt or repackaged on the way.
Secrets Generated For You
The 2 secrets this app needs are generated at launch and kept in your project's encrypted vault, never typed by hand or stored in plain settings.
Your Admin Sign-In, Ready
The first admin password is generated for you; you reveal it in the Secrets Vault (Guild plan and up).
Everything It Needs, Beside It
ollama/ollama:0.40.13 starts next to it on the same server, on a private network the internet cannot reach.
Your Data Stays On Your Server
Its data folders live on your server and are kept across every redeploy and update.
Its Address Set For You
The app is told its own public address at every deploy, on your custom domain with https or on the server's IP.
An Image Vulnerability Scan
At every deploy the image is scanned on your server for known vulnerabilities and the counts are shown in the deploy log. The scan reports; it does not block the deploy.
The Facts
- Publisher
- Open WebUI Inc. (chat) and Ollama (model runtime)
- Official image
ghcr.io/open-webui/open-webui:0.12- Runs beside it
- ollama/ollama:0.40.13
- Server memory
- 4 GB RAM or more
- Licence
- Open WebUI License (BSD-3-style, plus a clause that the Open WebUI branding may not be removed above 50 users a month without permission; not an OSI open-source licence). Ollama: MIT. DeepSeek-R1 1.5B and 7B: MIT; 8B: Llama 3.1 licence.
- Links
- Homepage · Documentation
Good To Know Before You Launch
- Sign in with your admin email and the password in this project's Secrets Vault.
- Download a model once: Admin Settings → Models → pull "deepseek-r1:1.5b" (about 1.1 GB). Larger models need more memory: 7B and 8B want an 8 GB server and answer a few words a second without a graphics card.
- Everything runs on your server's processor; there is no graphics card on an ordinary VPS.
- Add a custom domain: the publisher advises against plain http for real use.