Open WebUI

Open WebUI as a self-run interface for local and external models

Open WebUI is an open-source, self-hostable web interface for language models. It pools local models via Ollama and external, OpenAI-compatible interfaces in a single interface that can run entirely offline.

Anyone who does not want to use language models through a single provider's front end needs an interface of their own that pools several models and runs on their own infrastructure. Open WebUI is one such interface. It reproduces the feel of a commercial AI chat while leaving the model underneath open, and it puts the emphasis on operation without an internet dependency. This page describes what Open WebUI is, what it can do and under which licence it stands.

Interface under the operator's own control

Open WebUI is a self-hostable web application that its developers describe as an extensible AI platform built for fully local operation. The interface talks to two kinds of models. First, open-weights models run locally via Ollama, which work on the local machine without an external connection. Second, any service with an OpenAI-compatible interface, from a cloud API to a further local server. Both kinds appear side by side in the same interface, so Language Models can be compared and mixed without switching tools.

Open WebUI is installed via Docker, the Python package pip or Kubernetes. That makes it an interface under the operator's own control, comparable to LibreChat, but with a pronounced focus on local, Ollama-driven operation.

What Open WebUI provides

Beyond plain chat, Open WebUI brings broad equipment that makes for a productive model access point:

Local RAG. A built-in retrieval-augmented generation function allows querying internal documents; several vector databases are available as the store.
Web search. Requests can be enriched with a web search, with numerous search providers available for connection, among them Google, Brave Search, DuckDuckGo and Bing.
Multi-user operation. Accounts, user groups and role-based access control separate permissions instead of relying on a shared login.
Tools and functions. Custom tools and processing steps can be added through Python functions and pipelines; connecting external tools via the Model Context Protocol is possible through a proxy.
Image generation and voice. Image generation through services such as ComfyUI or AUTOMATIC1111, plus voice-driven operation through various speech-to-text and text-to-speech providers.
Model management. Custom Ollama models can be created and managed directly from the interface.

The common thread of these features is the claim to work fully offline. Anyone who uses only local models and local vector databases can run Open WebUI without any outbound connection. As soon as external APIs or a web search are wired in, the relevant part of the request leaves the local environment, a boundary that operators have to draw deliberately.

Under which licence Open WebUI stands

The licence of Open WebUI deserves a close look, because it differs from a purely permissive licence. Up to and including version 0.6.5 the project was under the BSD-3-Clause licence, a permissive licence that leaves use, modification and redistribution largely free. With version 0.6.6, effective from 19 April 2025, an additional condition was added: the Open WebUI branding, meaning name, logo and comparable identifiers, may not be removed, altered or obscured in the interface.

This branding condition does not apply in every case. It is waived for deployments with at most fifty end users within thirty days, with explicit prior written permission from the copyright holder, and under an enterprise licence that permits such a change. Code contributed before version 0.6.6 remains under BSD-3-Clause. In practice this means the source code is openly inspectable and the project is developed transparently, but because of the branding condition it is no longer an open-source licence in the narrower sense recognised by the Open Source Initiative. Anyone running Open WebUI in house should be aware of this condition before rebranding the interface or passing it on as part of a product of their own.

Where Open WebUI fits

Open WebUI is an interface, not a model. The value only emerges in combination with the models underneath, whether an open-weights model run locally via Ollama or an external API. Compared with LibreChat, Open WebUI pursues a similar goal, a self-operated interface in front of freely chosen models, but it puts a stronger emphasis on local operation and a broad feature set built into the interface. As soon as the interface calls tools and chains steps, it touches the field of AI agents, for which a chat interface can be the visible entry point. Which of the two interfaces fits depends on the use case, on the desired depth of features and on the willingness to observe the branding condition of the licence described above.

References

Open WebUI Licence text and licence history. Branding condition from version 0.6.6, exemptions for small deployments and the change from the BSD-3-Clause licence. (19.04.2025). docs.openwebui.com/license/
Open WebUI Project overview and feature list. Self-hostable AI platform for offline operation, Ollama integration, OpenAI-compatible interfaces, RAG, web search and multi-user operation. (2026). github.com/open-webui/open-webui
Open WebUI Documentation. Feature scope, supported model connections and installation paths via Docker, pip and Kubernetes. (2026). docs.openwebui.com/

Ask AI

These links open external AI services, the conversation and its content are sent to their providers.