Description

isModelNotFoundErrorMessage() in src/agents/live-model-errors.ts does not recognize the error format "Not supported model <model-id>" returned by some OpenAI-compatible providers (e.g. internal gateways that proxy multiple model backends).

This causes the error to fall through to isSchemaErrorMessage() → classified as "schema" → user sees:

LLM request failed: provider rejected the request schema or tool payload.

instead of the correct model-not-found handling.

Steps to reproduce

1. Configure an OpenAI-compatible provider with an invalid model ID
2. The provider returns HTTP 400 with body like:

   {"error":{"code":"400","message":"Param Incorrect","param":"Not supported model some-model-id"}}

3. The error is classified as "schema" instead of "model_not_found"

Expected behavior

The error should be classified as "model_not_found" and displayed as a model configuration issue, not a schema/tool payload rejection.

Root cause

isModelNotFoundErrorMessage() handles patterns like:

• "model not found"
• "unknown model"
• "invalid model"
• "model does not exist"

But NOT "Not supported model xxx" — a format used by some Chinese cloud provider gateways (Xiaomi MiFY, etc.).

Suggested fix

Add one line to src/agents/live-model-errors.ts:

if (/not supported model/i.test(msg)) {
  return true;
}

This pattern is unambiguous and won't false-positive on other error types.

Impact

Without this fix, any user on a provider that returns this error format will:

• See a misleading "schema rejected" error
• Waste time debugging tool schemas instead of checking their model config
• Have failover logic skip the model (format errors are non-retriable) instead of reporting model-not-found