[Feature] Support Advanced Model Configuration

[luokaiyue2035goals]

What do you need? / 你需要什么？

it is suggested to set a defualt model
建议可以设置默认模型

[zerob13]

Currently, DeepChat only allows a single default model configuration, which limits flexibility for different usage scenarios.
Building on the existing request [#1174 – set a default model], this feature proposes a more granular and extensible model configuration system, allowing users to assign different models to different responsibilities.

This would significantly improve performance, cost control, and extensibility for advanced users and agent-based workflows.

---

Proposed Model Configuration Types

1. Default Models (Core Usage)

Allow users to separately configure default models for different core use cases:

• Default Chat Model
  Used for standard user conversations.

• Default Agent Model
  Used for agent reasoning, planning, and task execution.

• Default ACP Agent Model
  Used specifically for ACP-based agents (if enabled).

This separation allows users to balance quality, speed, and cost based on actual usage patterns.

---

2. Assistant Model (Lightweight Helper Model)

Introduce a dedicated Assistant Model, used for auxiliary tasks such as:

• Conversation title generation
• Context summarization / compression
• Future memory extraction
• Tool selection and automation assistance

Recommendation:
This model is expected to be lightweight and fast (e.g. smaller or cheaper models), as it runs frequently but does not require high reasoning depth.

---

3. Visual Assistant Model (Optional)

Add support for a Visual Assistant Model, used to assist features such as:

• Vision-related MCP workflows
• Image understanding or recognition support
• Visual-enabled assistant capabilities

Configuration options:

• Select a vision-capable model
• Or set to none to explicitly disable visual assistance

This makes visual capabilities explicit and optional, rather than implicitly tied to the main model.

---

Benefits

• Clear separation of responsibilities between models
• Better performance and cost optimization
• More flexibility for agent-based and future automation features
• Cleaner architecture for extending multimodal and memory-related capabilities

---

Notes

This feature is designed to be backward-compatible:
If users do not configure these options explicitly, DeepChat can fall back to the existing single default model behavior.