Problem with connecting my instance of LLM

Help Agent Builder
1

Hello,

I’ve got a problem connecting my company’s instance of ChatGPT on Azure.

Usually, I would use the preexisting connector “OpenAI V1 Compliant LLM”, as I’m using an endpoint and API key.

The problem is that it requires the URL to have the endpoint “/models” - so it can check what models are deployed.

Due to company policy, the requests are routed through PowerProxy, and the link is not standard - I have a specific URL that I have to use:
https://powerproxy.mycompany.com/openai/deployments//chat/completions?api-version=

I found a way to create a custom connector to use this specific endpoint. I also tried and succeeded in using a simple HTTP request.

My issue and question is - How can I use my LLM as a model for Agents built in Studio Cloud? Is it even possible? If I can’t change the way we are connecting to the model, are there any workarounds?

Thanks in advance

2

Yes, it may be possible, but not through a plain custom connector alone. The key question is whether UiPath can successfully register and validate your model as an LLM provider. The /models discovery requirement is often the blocker.

From UiPath documentation, the OpenAI V1 Compliant LLM connector is designed to connect any model that follows the OpenAI V1 specification, and UiPath performs a validation/probe process before the model can be used by Agent Builder.

Ask the PowerProxy team to expose a compatible endpoint: GET /models
Therefore, the model is unlikely to be usable directly inside Studio Cloud Agents/Agent Builder in its current form. The most realistic enterprise-grade solution is either:

  1. Expose a compatible /models endpoint through PowerProxy, or
  2. Place a thin OpenAI-compatible wrapper/API Management layer in front of PowerProxy.