diff --git a/openadapt/config.py b/openadapt/config.py
deleted file mode 100644
index 3580e8f28..000000000
--- a/openadapt/config.py
+++ /dev/null
@@ -1,193 +0,0 @@
-"""Shared configuration for OpenAdapt packages.
-
-This module provides a unified configuration interface that loads settings
-from environment variables and .env files.
-
-Usage:
-    from openadapt.config import settings
-
-    # Access API keys
-    anthropic_key = settings.anthropic_api_key
-
-    # Access paths
-    capture_dir = settings.capture_dir
-
-    # Access model settings
-    default_model = settings.default_model
-"""
-
-from pathlib import Path
-from typing import Optional
-
-from pydantic_settings import BaseSettings, SettingsConfigDict
-
-
-class OpenAdaptSettings(BaseSettings):
-    """Unified settings for all OpenAdapt packages.
-
-    Settings can be configured via:
-    1. Environment variables (OPENADAPT_* prefix)
-    2. .env file in current directory
-    3. Default values
-
-    Environment variable names are case-insensitive and use underscores.
-    Example: OPENADAPT_CAPTURE_DIR or openadapt_capture_dir
-    """
-
-    model_config = SettingsConfigDict(
-        env_prefix="OPENADAPT_",
-        env_file=".env",
-        env_file_encoding="utf-8",
-        extra="ignore",  # Ignore unknown env vars
-    )
-
-    # =========================================================================
-    # API Keys (no prefix, for compatibility with other tools)
-    # =========================================================================
-    anthropic_api_key: Optional[str] = None
-    openai_api_key: Optional[str] = None
-    google_api_key: Optional[str] = None
-    lambda_api_key: Optional[str] = None
-
-    # =========================================================================
-    # Paths
-    # =========================================================================
-    # Default directory for capture recordings
-    capture_dir: Path = Path.home() / ".openadapt" / "captures"
-
-    # Default directory for training outputs
-    training_output_dir: Path = Path.home() / ".openadapt" / "training"
-
-    # Default directory for benchmark results
-    benchmark_results_dir: Path = Path.home() / ".openadapt" / "benchmarks"
-
-    # Model cache directory
-    model_cache_dir: Path = Path.home() / ".openadapt" / "models"
-
-    # Embedding cache directory
-    embedding_cache_dir: Path = Path.home() / ".openadapt" / "embeddings"
-
-    # =========================================================================
-    # ML Settings
-    # =========================================================================
-    # Default model for training/inference
-    default_model: str = "Qwen/Qwen3-VL-2B-Instruct"
-
-    # Default device for inference
-    default_device: str = "auto"  # "auto", "cuda", "mps", "cpu"
-
-    # Default batch size for training
-    default_batch_size: int = 1
-
-    # Default learning rate
-    default_learning_rate: float = 2e-5
-
-    # =========================================================================
-    # Capture Settings
-    # =========================================================================
-    # Enable audio recording by default
-    capture_audio: bool = True
-
-    # Enable Whisper transcription by default
-    capture_transcribe: bool = False
-
-    # Screenshot interval in seconds (0 for event-driven only)
-    capture_screenshot_interval: float = 0.0
-
-    # =========================================================================
-    # Evaluation Settings
-    # =========================================================================
-    # Default max steps for evaluation
-    eval_max_steps: int = 15
-
-    # Default number of tasks for mock evaluation
-    eval_default_tasks: int = 10
-
-    # Default evaluation agent
-    eval_default_agent: str = "api-claude"
-
-    # =========================================================================
-    # Server Settings
-    # =========================================================================
-    # Default server port
-    server_port: int = 8080
-
-    # Default server host
-    server_host: str = "127.0.0.1"
-
-    # Auto-open browser
-    server_auto_open: bool = True
-
-    # =========================================================================
-    # Azure Settings (for cloud evaluation)
-    # =========================================================================
-    azure_subscription_id: Optional[str] = None
-    azure_ml_resource_group: Optional[str] = None
-    azure_ml_workspace_name: Optional[str] = None
-    azure_docker_image: Optional[str] = None
-
-    # =========================================================================
-    # Grounding Settings
-    # =========================================================================
-    # OmniParser server URL
-    omniparser_url: Optional[str] = None
-
-    # UI-TARS server URL
-    uitars_url: Optional[str] = None
-
-    # =========================================================================
-    # Retrieval Settings
-    # =========================================================================
-    # Default embedding dimension
-    retrieval_embedding_dim: int = 512
-
-    # Default number of results
-    retrieval_top_k: int = 5
-
-    def ensure_directories(self) -> None:
-        """Create all configured directories if they don't exist."""
-        for path in [
-            self.capture_dir,
-            self.training_output_dir,
-            self.benchmark_results_dir,
-            self.model_cache_dir,
-            self.embedding_cache_dir,
-        ]:
-            path.mkdir(parents=True, exist_ok=True)
-
-    def get_device(self) -> str:
-        """Get the appropriate device for ML operations.
-
-        Returns:
-            Device string: "cuda", "mps", or "cpu"
-        """
-        if self.default_device != "auto":
-            return self.default_device
-
-        try:
-            import torch
-
-            if torch.cuda.is_available():
-                return "cuda"
-            elif hasattr(torch.backends, "mps") and torch.backends.mps.is_available():
-                return "mps"
-        except ImportError:
-            pass
-
-        return "cpu"
-
-
-# Global settings instance
-settings = OpenAdaptSettings()
-
-
-# Convenience function to reload settings (useful for testing)
-def reload_settings() -> OpenAdaptSettings:
-    """Reload settings from environment and .env file.
-
-    Returns:
-        New OpenAdaptSettings instance.
-    """
-    global settings
-    settings = OpenAdaptSettings()
-    return settings