Open WebUI
nigiri.open-webui
Self-hosted chat interface for local and remote language models, adding document retrieval, model switching, per-user workspaces, tool calling and an extensible pipeline layer.
A Docker Compose file for Open WebUI alone, generated from the catalog and checked by the safety inspector. Sign in and use Kitchen Credits to activate it.
Uses the Medusa cart and payment-session path. This test control does not create an order or capture payment.
Open WebUI provides the conversational front end that a bare model runtime lacks: threaded chats, per-user accounts and permissions, saved prompts, model presets and side-by-side comparison of two models on the same question. Documents uploaded to a workspace are chunked, embedded and retrieved as context, and web search can be attached to a conversation as an additional retrieval source. It speaks the OpenAI API format, so it connects equally to a local runtime and to hosted providers, often several at once. A pipelines layer lets custom Python run inside the request path for filtering, routing or tool execution.
You know it worked when
- The interface loads and the first account is created as administrator.
- A configured model backend appears in the model selector.
- A chat message returns a streamed response.
- A document uploaded to a workspace is retrieved as context in an answer.
- A second, non-administrator account sees only its own chats.
Known sharp edges
- The branding clause bites at 51 users, which is a licensing threshold with no technical signal — nothing in the software warns when a deployment crosses it.
- The image tag follows a development branch rather than a release, so an unattended pull can bring in breaking changes between restarts.
- The data directory holds accounts, chat history and the vector index together, and losing it destroys all three.
- Pipelines execute arbitrary Python inside the request path, so anyone able to install one has code execution in the container.