[Feature]: Recommendations for Tool-Using Coding Agents

### Problem or motivation

I would like to see llmfit provide recommendations not only for models intended for direct chat/prompt usage, but also for models that are suitable for tool-using coding agents, such as [OpenCode](https://opencode.ai/).
The memory requirements and practical performance characteristics can be quite different when a model is used through an agent that interacts with tools, maintains context, and performs multiple operations, compared with simply sending prompts to a model.

I have been experimenting with running models locally through LM Studio and OpenCode.
I tried the following models:

- DeepSeek-R1-Distill-Qwen-7B
- Qwen2.5-Coder-7B-Instruct

Although these models may appear suitable based on their model specifications and available memory, I encountered GPU memory errors when using them through OpenCode.
The errors I encountered were:
```
RuntimeError: [METAL] Command buffer execution failed:
Insufficient Memory
(00000008:kIOGPUCommandBufferCallbackErrorOutOfMemory)
```
I encountered the same type of error with both models.

### Proposed solution

It would be useful if llmfit could distinguish between different use cases when making recommendations for OpenCode or similar coding agents.

### Alternatives considered

_No response_

### Feature area

Provider integration (Ollama, llama.cpp, MLX, Docker, LM Studio)

### Would you be willing to contribute this?

I could help with guidance

### Additional context

_No response_

Start by reviewing the provider-integration paths for Ollama, llama.cpp, MLX, Docker, and LM Studio, along with how current recommendations are produced. Define how OpenCode or similar tool-using coding agents should be distinguished from direct prompt use, then verify that recommendations account for their memory and practical performance requirements.

Rédigé par le modèle d'indexation à partir du texte de l'issue.