Integration of Thinking Models (e.g., DeepSeek-R1)

[maciejmajek]

Is your feature request related to a problem? Please describe.
Standard models (versus thinking models) struggle with advanced planning, making complex task execution inefficient. Thinking models (e.g., DeepSeek-R1) demonstrate strong reasoning abilities, making them ideal for structured task execution in RAI. To fully leverage AI-driven automation, I need specialized models for different aspects of reasoning and execution.

Describe the solution you'd like
I want to integrate thinking models (e.g., DeepSeek-R1) into RAI using LangGraph-based architectures. Specifically, this setup would include:

• Thinking Model: A model dedicated to structuring and breaking down tasks before execution.
• Tool-Calling Model: A model optimized for executing predefined plans by calling relevant tools.
• Multimodal Tool-Calling Models: A hybrid system capable of handling both textual and multimodal inputs while executing planned actions.

LangGraph nodes should coordinate these models. Additionally, config.toml should be expanded to support the inclusion of these additional models, allowing for proper configuration and execution within the system.