## Summary

The LiteLLM Proxy UI "Ask AI" feature (`POST /usage/ai/chat`) fails when selecting a model alias defined in `model_list` if the alias is not itself a native LiteLLM provider-prefixed model string.

The rest of the proxy works with these aliases, but Ask AI appears to pass the selected alias directly to `litellm.acompletion()` instead of resolving it through the proxy router / model list first.

## Environment

- LiteLLM version: `1.82.6` - Deployment: LiteLLM proxy Docker image `docker.litellm.ai/berriai/litellm:main-latest` - `STORE_MODEL_IN_DB=True` - UI route involved: `POST /usage/ai/chat`

## Example model config

```yaml model_list: - model_name: minimax2openai/minimax-m2-7-highspeed litellm_params: model: MiniMax-M2.7-highspeed custom_llm_provider: openai api_base: os.environ/MINIMAX_OPENAI_API_BASE api_key: os.environ/MINIMAX_API_KEY model_info: input_cost_per_token: 0.0000006 output_cost_per_token: 0.0000024 ```

This alias is returned correctly by `GET /v2/model/info` and works through normal proxy completion endpoints.

## Reproduction

1. Configure a proxy model alias such as `minimax2openai/minimax-m2-7-highspee…