Open WebUI

Use Open WebUI as a chat front end for your Verdictan gateway. An administrator adds one OpenAI-compatible connection that points at the gateway, and every chat through it passes your policies.

The connection uses one Verdictan application token for all Open WebUI users. Verdictan sees that token, not the individual Open WebUI user.

Before you start

• A working gateway from the quickstart.
• A client API token for Open WebUI, created in the Console under API Tokens with the Access Key role.
• A network path from the Open WebUI server to the gateway. In a container, 127.0.0.1 is the container itself, not the gateway host.

Add the connection

1. In Open WebUI, open Settings → Admin → Connections.
2. Select + next to Manage OpenAI API Connections, and keep the OpenAI-compatible type.
3. Enter a gateway base URL ending in /v1:
   • Open WebUI on the gateway host, outside a container: http://127.0.0.1:41002/v1
   • Open WebUI in a container or on another host: your agent's central base URL, or another gateway address the server can reach
4. Enter the Verdictan application token as the API key, and save.

Open WebUI then loads the model list from the gateway's GET /v1/models, sending the token on that call and on each chat request.

This is a server-side connection, so the token stays on the Open WebUI server. Open WebUI's browser-side Direct Connections handle networking and credentials differently; see its connection guide.

Verify

1. Pick one of the gateway's models in Open WebUI and send Reply with Hello. This is an integration check.
2. Check that the assistant replies.
3. Find the request in Verdictan History by time and model (capture must be on), or in the delivered gateway events.

A filled model picker proves only that model discovery works. The reply plus a matching Verdictan record proves the chat went through the gateway. Test file uploads, retrieval, tools, and images separately before you turn them on for users.

Troubleshooting

Symptom	Cause	Fix
The model picker is empty	GET /v1/models failed	Check the base URL, the token, and the network path from the Open WebUI server.
Every event shows the same token	Open WebUI sends one credential for all users	Use Open WebUI's own records to see who sent what. To separate teams, create a token and a connection per team.
The reply arrives in one piece, with the model name verdictan-buffered-stream	The chain needs the complete output, so the gateway holds the stream	Nothing to fix. It is the complete, governed answer.
Open WebUI shows an error	Authentication, a policy, or the provider failed	Read the status and error code. For a policy block, see Investigate a Blocked Request.

Next steps

• verdictan events: query and export the events Open WebUI traffic creates.
• Configure Routes and Consumer Groups: give a team's token its own chain.