Skip to main content

Connect Open WebUI

Use Primex models in an existing Open WebUI installation. You need administrator access to its connection settings, a Primex API key and AI credit for chat requests. Choose a model in Open WebUI Setup.

Add the connection​

Open Settings → Admin → Connections, find Manage OpenAI API Connections, and click Add Connection.

SettingValue
Connection / providerOpenAI-compatible
URLhttps://gateway.primexos.com/v1
API KeyYour dedicated Primex key, entered privately
ModelThe exact ID selected in Setup

Save the connection. Use Verify Connection to check model discovery; Primex's /v1/models read requires a valid key and is non-billable. Then choose the model in a new chat. If needed, use the connection's Model IDs allowlist to expose only your chosen IDs. Refresh discovery when a model is withdrawn.

Saving or checking discovery is followed by an actual chat request. For remote installations, the Open WebUI server needs HTTPS access to the Primex Gateway. Keep the key in the connection credential; do not paste it into a conversation.

First conversation​

  1. Select the Primex model explicitly in a new chat.
  2. Leave attachments, retrieval and external tools off.
  3. Send “Reply with a short greeting”, then ask a follow-up referring to it.
  4. Confirm complete answers and the selected model in Usage.

For chat parameters, omit optional sampling controls initially, or use Temperature 1 where a value is required. Keep the output limit within the model's published bounds. Connection/provider retry behavior is separate from Primex routing; inspect Usage before repeating an interrupted request.

Documents, tools and additional models​

Chat Completions is the connection used here. Responses uses a different protocol and requires an eligible model and appropriate client configuration. Select embedding models separately for document ingestion/retrieval. Image input, tools and reasoning controls depend on the chosen offering; a connection does not enable those features automatically.

Fix a connection or remove it​

  • 401: check the connection's key and authentication mode.
  • 404 / model missing: use /v1 once, refresh discovery, and check the exact ID.
  • 400: remove unsupported parameters; check Temperature and model capabilities.
  • 402 / 429 / interrupted stream: inspect Errors & limits and Usage.

To disconnect, replace the model in dependent workspaces, remove the connection and revoke its dedicated key when no longer used.

See the official provider guide. Report a setup issue.