Open WebUI is a self-hosted web interface for talking to AI models. You run it on your own computer or server, connect it to a local runtime such as Ollama or to any OpenAI-compatible API, and get a ChatGPT-style workspace with document search, custom assistants and user accounts. This overview is based on the project’s own documentation; we have not run a hands-on test, so we make no claims about speed or quality in a particular setup.
What Open WebUI is
The documentation describes Open WebUI as a “self-hosted AI platform that’s extensible, feature-rich, user-friendly, and built to run entirely offline.” Its pitch is to replace the patchwork of separate tools a team might use for chat, image generation and document lookup with one interface. It supports Ollama and OpenAI-compatible APIs, and the features page says you can chat with Ollama, OpenAI, Anthropic or any OpenAI-compatible provider from the same screen, attach files, search the web and let the model use tools. You can switch models in the middle of a conversation.
Because it is only the interface, Open WebUI does not include a model by default. You bring one: a local model served by Ollama or LM Studio, or a hosted model through an API key. That separation is the main reason people use it, since the same front end can sit on top of whichever models you choose. Our earlier Tool of the Day on Ollama and LM Studio overview cover two common back ends.
Main features in the documentation
The features page lists several building blocks:
- Knowledge. Upload files and build knowledge bases, then choose between vector search (retrieval-augmented generation) for large collections or injecting full content for precision. Models with native function calling can search and synthesize across the collection.
- Custom models. A wrapper that combines a base model with a system prompt, knowledge and tools, such as a “Code Reviewer” with your lint rules. The docs call every such agent “a configuration wrapper.”
- Notes and Channels. A workspace for drafts outside any one chat, and shared spaces where team members and models share a conversation, tagged by name.
- Open Terminal. A connected computing environment where the model writes code, runs it, reads the output and iterates.
- Extensibility. Python tools, plus MCP or OpenAPI servers, can add capabilities. Community extensions can be imported.
- Access control. Roles, user groups, per-model permissions and identity-provider integration, described as multi-user “from day one.”
The docs also describe a stateless architecture meant to scale horizontally on Docker, Kubernetes, pip or bare metal, with shared databases and Redis coordinating instances.
Getting started
The quick start says Open WebUI runs on macOS, Linux (x86_64 and ARM64) and Windows. The documented Docker route is a single docker run command that publishes the interface on port 3000, stores data in a named volume and sets a fixed WEBUI_SECRET_KEY. The docs warn never to run without the volume, which holds chats, users and settings, and note that without a fixed key every recreated container logs everyone out.
Image variants matter for size. The standard :main image bundles speech-to-text and embedding models, while :slim is about 197 MB instead of 1.62 GB to download, according to the docs, which say they checked on October 16, 2026 for Linux/amd64. Slim leaves out the local machine-learning stack, so features like knowledge search, document extraction and voice need external services. There are also CUDA variants for Nvidia GPUs and an :ollama variant that bundles Ollama. A pip install open-webui route and a desktop app are also listed. If you want to learn Ollama first, see our step-by-step Ollama guide.
Licensing to check
Open WebUI is free to use, but its license is not plain BSD-3. The license page says that from version 0.6.6 (April 2025) it added a branding-protection clause. Branding requirements apply to deployments of more than 50 users in aggregate, while self-hosted internal deployments with 50 or fewer users may rebrand. Anyone wanting no restrictions can fork from version 0.6.5. Read the full license on GitHub before building a commercial product on it.
Who it suits
Open WebUI fits people who already run models locally and want a nicer front end, small teams that want a shared, private chat workspace, and developers who want to mix local and hosted models behind one login. It is less suited to someone who just wants a quick answer from a hosted assistant, since you must host and secure it yourself. If you expose it beyond your own machine, keep it updated and put it behind proper authentication.
What to watch next
- How the project’s separate “Computer” effort, aimed at real-machine agents, graduates features into Open WebUI.
- Changes to the license or its branding clause.
- MCP and tool-calling support across different local models.
- Enterprise plans, including long-term support, for organizations.
Key takeaways
- Open WebUI is a self-hosted interface that works with Ollama and OpenAI-compatible providers.
- It adds knowledge bases, custom assistants, tools and multi-user access.
- Docker is the documented route, with a 197 MB slim image for API-only setups.
- Its license adds a branding clause for deployments of more than 50 users.
- This summary comes from public documentation, not hands-on testing.
Frequently asked questions
Does Open WebUI include its own AI model?
No. It is an interface. You connect it to a local runtime such as Ollama or to a hosted provider with an API key.
Can it run offline?
The documentation describes it as built to run entirely offline, as long as the models and services you use are also local.
Is Open WebUI free?
The software is free to use, but the license includes a branding clause for larger deployments, so check the terms on the license page before commercial use.
