LLM Passthrough not working

I have set the LLM credentials to use OpenAI, however general knowledge test questions are not being passed to the LLM, I just the an "unable to reach entity" error for instance.

Could you advise how it is meant to work to push general questions off to the LLM?

Thanks for your interest in Sophia.

I've tested this extensively now, and the LLM fallback is working fine. What's happening is there's an error somewhere such as incorrect model slug / API key or maybe it can't connect to the internet, or something. There's a configuration or permission issue somewhere.

The issue is Sophia isn't relaying the error back to you and instead just says it can't understand your message. This has been fixed and will be included in the upcoming v2.2 upgrade which will hopefully be out later today.

Hope that helps, and let me know if I can be of further assistance.

Warm regards, Matt

Matt, Thanks for the reply. I thought that may be the case. How do you give Sophia permission to access the internet Is there a permission within the configuration?

Thanks Stuart

Try the latest upgrade, LLM fallback is working according to the reports I'm getting. Please let me know if not though.

I'm on 2.2.0 and not having any luck with LLM fallback.... Here my config: LLM Trigger: ask+AI Provider: Google Model Name: gemini-3.1-flash-lite API Key: XXXXXX

Payload:

{
  "license_id": "XXXXX",
  "mode": "",
  "messages": [
    {
      "text": "Ask AI how far is the sun from the earth?",
      "response": "unfortunately, Unknown intent HassError has occurred.",
      "intents": [
        {
          "intent": "HassError",
          "text": "it seems there's been a hiccup, HTTP error: 401.",
          "slots": {},
          "device_candidates": [],
          "response_type": "error",
          "success": [],
          "failed": [],
          "error": "Unknown intent HassError"
        }
      ]
    }
  ]
}

Notes:

• "mode": "" is empty in the payload — not sure if that's expected or a symptom of the misconfiguration.
• The Home Assistant logs also show homeassistant.helpers.intent.UnknownIntent: Unknown intent HassError — it looks like whatever handles the 401 tries to surface it via a synthetic HassError intent that HA's intent system doesn't recognize, so the real error (the 401) gets masked by a second, unrelated error.

Any pointers on where the 401 is actually coming from (Sophia's own backend vs. the Google API call)?