Open WebUI

License	FOSS problems	Maintenance	SSO	Users	Tested
Open WebUI LicenseSource available license. Permissive-in-practice, but the 'Open WebUI' branding must stay visible past 50 users in any rolling 30 days, unless a paid enterprise license removes it.	—	Active	✅	Unlimited	✅

Description

Open WebUI is a self-hosted chat interface for large language models — a ChatGPT-style UI running on your own infrastructure, against models you choose. It connects to locally running models via Ollama, or to any OpenAI-compatible API.

Features

• Model-agnostic — one interface over local models (via Ollama) or any OpenAI-compatible API, swappable without changing how anyone interacts with it.
• Centralised access — a shared tool for members to use AI without individual accounts on an external provider, managed and billed in one place instead of scattered per-seat subscriptions.
• Self-hosted conversation history — chats stay on this server rather than a vendor's — though see License and governance for what "self-hosted" actually requires here.

Usage patterns

• Shared gateway to external APIs — the realistic pattern without dedicated hardware: Open WebUI is just the interface, and without a GPU (or heavy CPU inference) running models locally via Ollama isn't practical. Proxying to an external API reintroduces the third-party dependency this stack otherwise avoids, but centralizes and manages it rather than leaving it to individual member accounts.
• Local inference — genuinely self-contained only with real hardware behind it, a GPU or substantial RAM for CPU inference — not what this deployment currently has.

License and governance

Was plain BSD-3-Clause; moved to the "Open WebUI License" in April 2025 (v0.6.6+), adding one condition — the "Open WebUI" name and branding can't be removed or obscured. That requirement doesn't bind small deployments, though: it only applies past 50 end users in any rolling 30-day window, a threshold this deployment sits well under. Everything contributed through v0.6.5 stays under the original BSD-3-Clause regardless.

Conclusion

Why you would — a single, centrally managed interface to AI models, avoiding per-member accounts on an external provider and keeping conversation history on infrastructure this project controls.

Why you wouldn't — without a GPU or serious CPU inference hardware, it's a UI in front of an external API, not a self-contained AI setup — the actual privacy/independence benefit is smaller than "self-hosted" suggests until real hardware backs it.

Setup

1. Add DNS records pointing ai.<domain> to the server:

   • A record: ai → <ipv4>
   • AAAA record: ai → <ipv6>

2. In Coolify, go to New Resource → Service, search for Open WebUI, set the domain on the open-webui container (inside the container settings), then deploy.

3. Open https://ai.<domain> and create the admin account.