Open WebUI for self-hosted model chat

How it works

The web application manages conversations and connects to inference endpoints. Its knowledge and tool features add retrieval and actions at the interface layer, while the endpoint supplies the model.

What you need

You can deploy the UI separately from the inference server. For a shared instance, identify where user accounts, uploaded files and chat history will persist before adding more people.

What to try first

Connect a single endpoint and test account access with two users before importing a shared document. Ask each user for a quotation and inspect which files they can retrieve.

Try an exported conversation and a fresh login after a restart. Conversation history, attached files and the model service can have different persistence settings.

Where to look for failures

In a container, localhost refers to that container. Set the reachable endpoint explicitly, and review the permissions and execution behavior of any community functions you install.

Related Franklin pages

Use the same two or three documents in each workspace. Comparing source passages, index updates and data flow reveals differences that a fluent chat answer can hide.

• Try desktop model chat and search an indexed folder of your own documents. GPT4All for desktop document chat
• Set up separate chat workspaces around document collections and chosen model providers. AnythingLLM for document workspaces

For more on how model size affects deployment decisions, read Franklin’s guide. Small language models

Project documentation

• Current project README
• README at the reviewed revision