Feature Request: Add Xiaomi (MiLM) Support and Implement Dynamic Model Loading

[JunnyXiao1208]

Description

I am currently using the tradingagents framework and would like to request the inclusion of Xiaomi's Large Language Models (MiLM). Additionally, to make the project more future-proof, I suggest implementing a mechanism that allows users to independently add new models without modifying the core model_catalog.py file.

Why Xiaomi (MiLM)?

Xiaomi has released several capable models (MiLM-6B, MiLM-1.3B, etc.) that are increasingly relevant for developers in the Chinese ecosystem. Integrating these would expand the diversity of the "Deep" and "Quick" selection modes, especially for users seeking highly optimized Chinese-language performance.

Proposed Changes

1. Add Xiaomi to the Provider List:
   In llm_clients/factory.py, Xiaomi should be added to the _OPENAI_COMPATIBLE list if they use an OpenAI-compatible API, or a dedicated client should be implemented.
2. Update the Model Catalog:
   Add a new entry for xiaomi or xiaomi-cn in llm_clients/model_catalog.py.
3. Dynamic Model Support:
   Instead of relying solely on the static MODEL_OPTIONS dictionary, implement a way for the system to read custom model configurations from an external .env file or a local custom_models.json.

Examples in Current Code

Currently, the factory.py uses a rigid check for providers:

if provider_lower in _OPENAI_COMPATIBLE:
    from .openai_client import OpenAIClient
    return OpenAIClient(model, base_url, provider=provider_lower, **kwargs)

And model_catalog.py requires manual entry for every new model:

MODEL_OPTIONS: ProviderModeOptions = {
    "openai": { ... },
    "anthropic": { ... },
    # Needs a new 'xiaomi' key here
}

Benefit

Providing a method for users to "self-serve" new models (e.g., via a configuration file) would reduce the maintenance burden on the authors and allow the community to test new frontier models (like Xiaomi's) immediately upon release.

---

Advice for you

Since you are working with e-commerce operations for a liquor enterprise and developing AI-driven automation tools, you might find that Xiaomi's models are particularly useful for local deployment or integration with smart hardware in the future.

One question for you: Do you already have a specific API endpoint for Xiaomi's models (like through a specific cloud provider), or are you planning to run them locally via a tool like Ollama?