- **Bootstrap IoC-based architecture with modular services.**

- **Implement `MlxModelService` for local LLM backend.**
- **Introduce `DatabaseService` for MySQL integration.**
- **Add `HistoryService` to manage conversation context.**
- **Set up CLI interface via `CliUiService`.**
- **Establish EventBus for token streaming.**
- **Include conversation repository for data persistence.**
- **Add environment-based configuration management.**
- **Draft IoC architectural plan.**

services/model/base.py

@@ -0,0 +1,18 @@
+from abc import ABC, abstractmethod
+from typing import Iterator
+
+
+class AbstractModelService(ABC):
+    """LLM 백엔드 Strategy 인터페이스."""
+
+    @abstractmethod
+    def load(self) -> None:
+        """모델을 메모리에 로드한다."""
+
+    @abstractmethod
+    def stream(self, prompt: str, max_tokens: int) -> Iterator[str]:
+        """프롬프트를 받아 토큰을 스트리밍한다."""
+
+    @abstractmethod
+    def build_prompt(self, history: list[dict]) -> str:
+        """대화 히스토리를 모델 입력 형식으로 변환한다."""

services/model/mlx_model.py

@@ -0,0 +1,29 @@
+from typing import Iterator
+
+from services.model.base import AbstractModelService
+
+
+class MlxModelService(AbstractModelService):
+    """MLX 기반 로컬 LLM Strategy 구현체."""
+
+    def __init__(self, model_id: str):
+        self._model_id = model_id
+        self._model = None
+        self._tokenizer = None
+
+    def load(self) -> None:
+        from mlx_lm import load
+        print(f"모델 로딩 중: {self._model_id}")
+        self._model, self._tokenizer = load(self._model_id)
+
+    def build_prompt(self, history: list[dict]) -> str:
+        return self._tokenizer.apply_chat_template(
+            history,
+            tokenize=False,
+            add_generation_prompt=True,
+        )
+
+    def stream(self, prompt: str, max_tokens: int) -> Iterator[str]:
+        from mlx_lm import stream_generate
+        for chunk in stream_generate(self._model, self._tokenizer, prompt=prompt, max_tokens=max_tokens):
+            yield chunk.text