massgen 0.0.3__py3-none-any.whl → 0.1.0__py3-none-any.whl

massgen/cli.py

@@ -1,4 +1,5 @@
 #!/usr/bin/env python3
+# -*- coding: utf-8 -*-
 """
 MassGen Command Line Interface
 
@@ -16,34 +17,55 @@ Usage examples:
     python -m massgen.cli --config config.yaml
 
     # Multiple agents from config
-    python -m massgen.cli --config multi_agent.yaml "Compare different approaches to renewable energy"
+    python -m massgen.cli --config multi_agent.yaml "Compare different approaches to renewable energy"  # noqa
 """
 
 import argparse
 import asyncio
 import json
 import os
+import shutil
 import sys
-import yaml
+from datetime import datetime
 from pathlib import Path
-from typing import Dict, Any, Optional, List
-
-from .utils import MODEL_MAPPINGS, get_backend_type_from_model
-
+from typing import Any, Dict, List, Optional
 
-# Load environment variables from .env file
+import yaml
+from dotenv import load_dotenv
+from rich.console import Console
+from rich.panel import Panel
+from rich.table import Table
+
+from .agent_config import AgentConfig, TimeoutConfig
+from .backend.azure_openai import AzureOpenAIBackend
+from .backend.chat_completions import ChatCompletionsBackend
+from .backend.claude import ClaudeBackend
+from .backend.claude_code import ClaudeCodeBackend
+from .backend.gemini import GeminiBackend
+from .backend.grok import GrokBackend
+from .backend.inference import InferenceBackend
+from .backend.lmstudio import LMStudioBackend
+from .backend.response import ResponseBackend
+from .chat_agent import ConfigurableAgent, SingleAgent
+from .frontend.coordination_ui import CoordinationUI
+from .logger_config import _DEBUG_MODE, logger, setup_logging
+from .orchestrator import Orchestrator
+from .utils import get_backend_type_from_model
+
+
+# Load environment variables from .env files
 def load_env_file():
-    """Load environment variables from .env file if it exists."""
-    env_file = Path(".env")
-    if env_file.exists():
-        with open(env_file, "r") as f:
-            for line in f:
-                line = line.strip()
-                if line and not line.startswith("#") and "=" in line:
-                    key, value = line.split("=", 1)
-                    # Remove quotes if present
-                    value = value.strip("\"'")
-                    os.environ[key] = value
+    """Load environment variables from .env files.
+
+    Search order (later files override earlier ones):
+    1. MassGen package .env (development fallback)
+    2. User home ~/.massgen/.env (global user config)
+    3. Current directory .env (project-specific, highest priority)
+    """
+    # Load in priority order (later overrides earlier)
+    load_dotenv(Path(__file__).parent / ".env")  # Package fallback
+    load_dotenv(Path.home() / ".massgen" / ".env")  # User global
+    load_dotenv()  # Current directory (highest priority)
 
 
 # Load .env file at module import
@@ -53,15 +75,6 @@ load_env_file()
 project_root = Path(__file__).parent.parent.parent.parent
 sys.path.insert(0, str(project_root))
 
-from massgen.backend.response import ResponseBackend
-from massgen.backend.grok import GrokBackend
-from massgen.backend.claude import ClaudeBackend
-from massgen.backend.gemini import GeminiBackend
-from massgen.chat_agent import SingleAgent, ConfigurableAgent
-from massgen.agent_config import AgentConfig
-from massgen.orchestrator import Orchestrator
-from massgen.frontend.coordination_ui import CoordinationUI
-
 # Color constants for terminal output
 BRIGHT_CYAN = "\033[96m"
 BRIGHT_BLUE = "\033[94m"
@@ -77,22 +90,153 @@ BOLD = "\033[1m"
 class ConfigurationError(Exception):
     """Configuration error for CLI."""
 
-    pass
+
+def _substitute_variables(obj: Any, variables: Dict[str, str]) -> Any:
+    """Recursively substitute ${var} references in config with actual values.
+
+    Args:
+        obj: Config object (dict, list, str, or other)
+        variables: Dict of variable names to values
+
+    Returns:
+        Config object with variables substituted
+    """
+    if isinstance(obj, dict):
+        return {k: _substitute_variables(v, variables) for k, v in obj.items()}
+    elif isinstance(obj, list):
+        return [_substitute_variables(item, variables) for item in obj]
+    elif isinstance(obj, str):
+        # Replace ${var} with value
+        result = obj
+        for var_name, var_value in variables.items():
+            result = result.replace(f"${{{var_name}}}", var_value)
+        return result
+    else:
+        return obj
+
+
+def resolve_config_path(config_arg: Optional[str]) -> Optional[Path]:
+    """Resolve config file with flexible syntax.
+
+    Priority order:
+
+    **If --config flag provided (highest priority):**
+    1. @examples/NAME → Package examples (search configs directory)
+    2. Absolute/relative paths (exact path as specified)
+    3. Named configs in ~/.config/massgen/agents/
+
+    **If NO --config flag (auto-discovery):**
+    1. .massgen/config.yaml (project-level config in current directory)
+    2. ~/.config/massgen/config.yaml (global default config)
+    3. None → trigger config builder
+
+    Args:
+        config_arg: Config argument from --config flag (can be @examples/NAME, path, or None)
+
+    Returns:
+        Path to config file, or None if config builder should run
+
+    Raises:
+        ConfigurationError: If config file not found
+    """
+    # Check for default configs if no config_arg provided
+    if not config_arg:
+        # Priority 1: Project-level config (.massgen/config.yaml in current directory)
+        project_config = Path.cwd() / ".massgen" / "config.yaml"
+        if project_config.exists():
+            return project_config
+
+        # Priority 2: Global default config
+        global_config = Path.home() / ".config/massgen/config.yaml"
+        if global_config.exists():
+            return global_config
+
+        return None  # Trigger builder
+
+    # Handle @examples/ prefix - search in package configs
+    if config_arg.startswith("@examples/"):
+        name = config_arg[10:]  # Remove '@examples/' prefix
+        try:
+            from importlib.resources import files
+
+            configs_root = files("massgen") / "configs"
+
+            # Search recursively for matching name
+            # Try to find by filename stem match
+            for config_file in configs_root.rglob("*.yaml"):
+                # Check if name matches the file stem or is contained in the path
+                if name in config_file.name or name in str(config_file):
+                    return Path(str(config_file))
+
+            raise ConfigurationError(
+                f"Config '{config_arg}' not found in package.\n" f"Use --list-examples to see available configs.",
+            )
+        except Exception as e:
+            if isinstance(e, ConfigurationError):
+                raise
+            raise ConfigurationError(f"Error loading package config: {e}")
+
+    # Try as regular path (absolute or relative)
+    path = Path(config_arg).expanduser()
+    if path.exists():
+        return path
+
+    # Try in user config directory (~/.config/massgen/agents/)
+    user_agents_dir = Path.home() / ".config/massgen/agents"
+    user_config = user_agents_dir / f"{config_arg}.yaml"
+    if user_config.exists():
+        return user_config
+
+    # Also try with .yaml extension if not provided
+    if not config_arg.endswith((".yaml", ".yml")):
+        user_config_with_ext = user_agents_dir / f"{config_arg}.yaml"
+        if user_config_with_ext.exists():
+            return user_config_with_ext
+
+    # Config not found anywhere
+    raise ConfigurationError(
+        f"Configuration file not found: {config_arg}\n"
+        f"Searched in:\n"
+        f"  - Current directory: {Path.cwd() / config_arg}\n"
+        f"  - User configs: {user_agents_dir / config_arg}.yaml\n"
+        f"Use --list-examples to see available package configs.",
+    )
 
 
 def load_config_file(config_path: str) -> Dict[str, Any]:
-    """Load configuration from YAML or JSON file."""
+    """Load configuration from YAML or JSON file.
+
+    Search order:
+    1. Exact path as provided (absolute or relative to CWD)
+    2. If just a filename, search in package's configs/ directory
+    3. If a relative path, also try within package's configs/ directory
+
+    Supports variable substitution: ${cwd} in any string will be replaced with the agent's cwd value.
+    """
     path = Path(config_path)
 
-    # If file doesn't exist in current path, try massgen/configs/ directory
-    if not path.exists():
-        # Try in massgen/configs/ directory
-        configs_path = Path(__file__).parent / "configs" / path.name
-        if configs_path.exists():
-            path = configs_path
+    # Try the path as-is first (handles absolute paths and relative to CWD)
+    if path.exists():
+        pass  # Use this path
+    elif path.is_absolute():
+        # Absolute path that doesn't exist
+        raise ConfigurationError(f"Configuration file not found: {config_path}")
+    else:
+        # Relative path or just filename - search in package configs
+        package_configs_dir = Path(__file__).parent / "configs"
+
+        # Try 1: Just the filename in package configs root
+        candidate1 = package_configs_dir / path.name
+        # Try 2: The full relative path within package configs
+        candidate2 = package_configs_dir / path
+
+        if candidate1.exists():
+            path = candidate1
+        elif candidate2.exists():
+            path = candidate2
         else:
             raise ConfigurationError(
-                f"Configuration file not found: {config_path} (also checked {configs_path})"
+                f"Configuration file not found: {config_path}\n" f"Searched in:\n" f"  - {Path.cwd() / config_path}\n" f"  - {candidate1}\n" f"  - {candidate2}",
             )
 
     try:
@@ -102,186 +246,564 @@ def load_config_file(config_path: str) -> Dict[str, Any]:
             elif path.suffix.lower() == ".json":
                 return json.load(f)
             else:
-                raise ConfigurationError(
-                    f"Unsupported config file format: {path.suffix}"
-                )
+                raise ConfigurationError(f"Unsupported config file format: {path.suffix}")
     except Exception as e:
         raise ConfigurationError(f"Error reading config file: {e}")
 
 
 def create_backend(backend_type: str, **kwargs) -> Any:
-    """Create backend instance from type and parameters."""
+    """Create backend instance from type and parameters.
+
+    Supported backend types:
+    - openai: OpenAI API (requires OPENAI_API_KEY)
+    - grok: xAI Grok (requires XAI_API_KEY)
+    - sglang: SGLang inference server (local)
+    - claude: Anthropic Claude (requires ANTHROPIC_API_KEY)
+    - gemini: Google Gemini (requires GOOGLE_API_KEY or GEMINI_API_KEY)
+    - chatcompletion: OpenAI-compatible providers (auto-detects API key based on base_url)
+
+    Supported backend with external dependencies:
+    - ag2/autogen: AG2 (AutoGen) framework agents
+
+    For chatcompletion backend, the following providers are auto-detected:
+    - Cerebras AI (cerebras.ai) -> CEREBRAS_API_KEY
+    - Together AI (together.ai/together.xyz) -> TOGETHER_API_KEY
+    - Fireworks AI (fireworks.ai) -> FIREWORKS_API_KEY
+    - Groq (groq.com) -> GROQ_API_KEY
+    - Nebius AI Studio (studio.nebius.ai) -> NEBIUS_API_KEY
+    - OpenRouter (openrouter.ai) -> OPENROUTER_API_KEY
+    - POE (poe.com) -> POE_API_KEY
+    - Qwen (dashscope.aliyuncs.com) -> QWEN_API_KEY
+
+    External agent frameworks are supported via the adapter registry.
+    """
     backend_type = backend_type.lower()
 
+    # Check if this is a framework/adapter type
+    from massgen.adapters import adapter_registry
+
+    if backend_type in adapter_registry:
+        # Use ExternalAgentBackend for all registered adapter types
+        from massgen.backend.external import ExternalAgentBackend
+
+        return ExternalAgentBackend(adapter_type=backend_type, **kwargs)
+
     if backend_type == "openai":
         api_key = kwargs.get("api_key") or os.getenv("OPENAI_API_KEY")
         if not api_key:
             raise ConfigurationError(
-                "OpenAI API key not found. Set OPENAI_API_KEY or provide in config."
+                "OpenAI API key not found. Set OPENAI_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
             )
-        return ResponseBackend(api_key=api_key)
+        return ResponseBackend(api_key=api_key, **kwargs)
 
     elif backend_type == "grok":
         api_key = kwargs.get("api_key") or os.getenv("XAI_API_KEY")
         if not api_key:
             raise ConfigurationError(
-                "Grok API key not found. Set XAI_API_KEY or provide in config."
+                "Grok API key not found. Set XAI_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
             )
-        return GrokBackend(api_key=api_key)
+        return GrokBackend(api_key=api_key, **kwargs)
 
     elif backend_type == "claude":
         api_key = kwargs.get("api_key") or os.getenv("ANTHROPIC_API_KEY")
         if not api_key:
             raise ConfigurationError(
-                "Claude API key not found. Set ANTHROPIC_API_KEY or provide in config."
+                "Claude API key not found. Set ANTHROPIC_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
             )
-        return ClaudeBackend(api_key=api_key)
+        return ClaudeBackend(api_key=api_key, **kwargs)
 
     elif backend_type == "gemini":
-        api_key = (
-            kwargs.get("api_key")
-            or os.getenv("GOOGLE_API_KEY")
-            or os.getenv("GEMINI_API_KEY")
-        )
+        api_key = kwargs.get("api_key") or os.getenv("GOOGLE_API_KEY") or os.getenv("GEMINI_API_KEY")
         if not api_key:
             raise ConfigurationError(
-                "Gemini API key not found. Set GOOGLE_API_KEY or provide in config."
+                "Gemini API key not found. Set GOOGLE_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
             )
-        return GeminiBackend(api_key=api_key)
+        return GeminiBackend(api_key=api_key, **kwargs)
+
+    elif backend_type == "chatcompletion":
+        api_key = kwargs.get("api_key")
+        base_url = kwargs.get("base_url")
+
+        # Determine API key based on base URL if not explicitly provided
+        if not api_key:
+            if base_url and "cerebras.ai" in base_url:
+                api_key = os.getenv("CEREBRAS_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Cerebras AI API key not found. Set CEREBRAS_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "together.xyz" in base_url:
+                api_key = os.getenv("TOGETHER_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Together AI API key not found. Set TOGETHER_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "fireworks.ai" in base_url:
+                api_key = os.getenv("FIREWORKS_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Fireworks AI API key not found. Set FIREWORKS_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "groq.com" in base_url:
+                api_key = os.getenv("GROQ_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Groq API key not found. Set GROQ_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "nebius.com" in base_url:
+                api_key = os.getenv("NEBIUS_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Nebius AI Studio API key not found. Set NEBIUS_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "openrouter.ai" in base_url:
+                api_key = os.getenv("OPENROUTER_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "OpenRouter API key not found. Set OPENROUTER_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and ("z.ai" in base_url or "bigmodel.cn" in base_url):
+                api_key = os.getenv("ZAI_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "ZAI API key not found. Set ZAI_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and ("moonshot.ai" in base_url or "moonshot.cn" in base_url):
+                api_key = os.getenv("MOONSHOT_API_KEY") or os.getenv("KIMI_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Kimi/Moonshot API key not found. Set MOONSHOT_API_KEY or KIMI_API_KEY environment variable.\n"
+                        "You can add it to a .env file in:\n"
+                        "  - Current directory: .env\n"
+                        "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "poe.com" in base_url:
+                api_key = os.getenv("POE_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "POE API key not found. Set POE_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
+                    )
+            elif base_url and "aliyuncs.com" in base_url:
+                api_key = os.getenv("QWEN_API_KEY")
+                if not api_key:
+                    raise ConfigurationError(
+                        "Qwen API key not found. Set QWEN_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
+                    )
+
+        return ChatCompletionsBackend(api_key=api_key, **kwargs)
+
+    elif backend_type == "zai":
+        # ZAI (Zhipu.ai) uses OpenAI-compatible Chat Completions at a custom base_url
+        # Supports both global (z.ai) and China (bigmodel.cn) endpoints
+        api_key = kwargs.get("api_key") or os.getenv("ZAI_API_KEY")
+        if not api_key:
+            raise ConfigurationError(
+                "ZAI API key not found. Set ZAI_API_KEY environment variable.\n" "You can add it to a .env file in:\n" "  - Current directory: .env\n" "  - Global config: ~/.massgen/.env",
+            )
+        return ChatCompletionsBackend(api_key=api_key, **kwargs)
+
+    elif backend_type == "lmstudio":
+        # LM Studio local server (OpenAI-compatible). Defaults handled by backend.
+        return LMStudioBackend(**kwargs)
+
+    elif backend_type == "vllm":
+        # vLLM local server (OpenAI-compatible). Defaults handled by backend.
+        return InferenceBackend(backend_type="vllm", **kwargs)
+
+    elif backend_type == "sglang":
+        # SGLang local server (OpenAI-compatible). Defaults handled by backend.
+        return InferenceBackend(backend_type="sglang", **kwargs)
+
+    elif backend_type == "claude_code":
+        # ClaudeCodeBackend using claude-code-sdk-python
+        # Authentication handled by backend (API key or subscription)
+
+        # Validate claude-code-sdk availability
+        try:
+            pass
+        except ImportError:
+            raise ConfigurationError("claude-code-sdk not found. Install with: pip install claude-code-sdk")
+
+        return ClaudeCodeBackend(**kwargs)
+
+    elif backend_type == "azure_openai":
+        api_key = kwargs.get("api_key") or os.getenv("AZURE_OPENAI_API_KEY")
+        endpoint = kwargs.get("base_url") or os.getenv("AZURE_OPENAI_ENDPOINT")
+        if not api_key:
+            raise ConfigurationError("Azure OpenAI API key not found. Set AZURE_OPENAI_API_KEY or provide in config.")
+        if not endpoint:
+            raise ConfigurationError("Azure OpenAI endpoint not found. Set AZURE_OPENAI_ENDPOINT or provide base_url in config.")
+        return AzureOpenAIBackend(**kwargs)
 
     else:
         raise ConfigurationError(f"Unsupported backend type: {backend_type}")
 
 
-def create_agents_from_config(config: Dict[str, Any]) -> Dict[str, ConfigurableAgent]:
+def create_agents_from_config(config: Dict[str, Any], orchestrator_config: Optional[Dict[str, Any]] = None) -> Dict[str, ConfigurableAgent]:
     """Create agents from configuration."""
     agents = {}
 
-    # Handle single agent configuration
-    if "agent" in config:
-        agent_config_data = config["agent"]
-        backend_config = agent_config_data.get("backend", {})
+    agent_entries = [config["agent"]] if "agent" in config else config.get("agents", None)
+
+    if not agent_entries:
+        raise ConfigurationError("Configuration must contain either 'agent' or 'agents' section")
+
+    for i, agent_data in enumerate(agent_entries, start=1):
+        backend_config = agent_data.get("backend", {})
+
+        # Substitute variables like ${cwd} in backend config
+        if "cwd" in backend_config:
+            variables = {"cwd": backend_config["cwd"]}
+            backend_config = _substitute_variables(backend_config, variables)
 
         # Infer backend type from model if not explicitly provided
-        if "type" not in backend_config and "model" in backend_config:
-            backend_type = get_backend_type_from_model(backend_config["model"])
-        else:
-            backend_type = backend_config.get("type")
-            if not backend_type:
-                raise ConfigurationError(
-                    "Backend type must be specified or inferrable from model"
-                )
+        backend_type = backend_config.get("type") or (get_backend_type_from_model(backend_config["model"]) if "model" in backend_config else None)
+        if not backend_type:
+            raise ConfigurationError("Backend type must be specified or inferrable from model")
+
+        # Add orchestrator context for filesystem setup if available
+        if orchestrator_config:
+            if "agent_temporary_workspace" in orchestrator_config:
+                backend_config["agent_temporary_workspace"] = orchestrator_config["agent_temporary_workspace"]
+            # Add orchestrator-level context_paths to all agents
+            if "context_paths" in orchestrator_config:
+                # Merge orchestrator context_paths with agent-specific ones
+                agent_context_paths = backend_config.get("context_paths", [])
+                orchestrator_context_paths = orchestrator_config["context_paths"]
+
+                # Deduplicate paths - orchestrator paths take precedence
+                merged_paths = orchestrator_context_paths.copy()
+                orchestrator_paths_set = {path.get("path") for path in orchestrator_context_paths}
+
+                for agent_path in agent_context_paths:
+                    if agent_path.get("path") not in orchestrator_paths_set:
+                        merged_paths.append(agent_path)
+
+                backend_config["context_paths"] = merged_paths
 
         backend = create_backend(backend_type, **backend_config)
-
-        # Create proper AgentConfig with backend_params
-        if backend_type.lower() == "openai":
-            agent_config = AgentConfig.create_openai_config(
-                **{k: v for k, v in backend_config.items() if k != "type"}
-            )
-        elif backend_type.lower() == "claude":
-            agent_config = AgentConfig.create_claude_config(
-                **{k: v for k, v in backend_config.items() if k != "type"}
-            )
-        elif backend_type.lower() == "grok":
-            agent_config = AgentConfig.create_grok_config(
-                **{k: v for k, v in backend_config.items() if k != "type"}
-            )
-        elif backend_type.lower() == "gemini":
-            agent_config = AgentConfig.create_gemini_config(
-                **{k: v for k, v in backend_config.items() if k != "type"}
-            )
+        backend_params = {k: v for k, v in backend_config.items() if k != "type"}
+
+        backend_type_lower = backend_type.lower()
+        if backend_type_lower == "openai":
+            agent_config = AgentConfig.create_openai_config(**backend_params)
+        elif backend_type_lower == "claude":
+            agent_config = AgentConfig.create_claude_config(**backend_params)
+        elif backend_type_lower == "grok":
+            agent_config = AgentConfig.create_grok_config(**backend_params)
+        elif backend_type_lower == "gemini":
+            agent_config = AgentConfig.create_gemini_config(**backend_params)
+        elif backend_type_lower == "zai":
+            agent_config = AgentConfig.create_zai_config(**backend_params)
+        elif backend_type_lower == "chatcompletion":
+            agent_config = AgentConfig.create_chatcompletion_config(**backend_params)
+        elif backend_type_lower == "lmstudio":
+            agent_config = AgentConfig.create_lmstudio_config(**backend_params)
+        elif backend_type_lower == "vllm":
+            agent_config = AgentConfig.create_vllm_config(**backend_params)
+        elif backend_type_lower == "sglang":
+            agent_config = AgentConfig.create_sglang_config(**backend_params)
         else:
-            # Fallback to basic config
             agent_config = AgentConfig(backend_params=backend_config)
 
-        # Set agent ID and system message
-        agent_config.agent_id = agent_config_data.get("id", "agent1")
-        agent_config.custom_system_instruction = agent_config_data.get("system_message")
-
-        agent = ConfigurableAgent(config=agent_config, backend=backend)
-        agents[agent.agent_id] = agent
-
-    # Handle multiple agents configuration
-    elif "agents" in config:
-        for agent_config_data in config["agents"]:
-            backend_config = agent_config_data.get("backend", {})
-
-            # Infer backend type from model if not explicitly provided
-            if "type" not in backend_config and "model" in backend_config:
-                backend_type = get_backend_type_from_model(backend_config["model"])
-            else:
-                backend_type = backend_config.get("type")
-                if not backend_type:
-                    raise ConfigurationError(
-                        "Backend type must be specified or inferrable from model"
-                    )
+        agent_config.agent_id = agent_data.get("id", f"agent{i}")
 
-            backend = create_backend(backend_type, **backend_config)
-
-            # Create proper AgentConfig with backend_params
-            if backend_type.lower() == "openai":
-                agent_config = AgentConfig.create_openai_config(
-                    **{k: v for k, v in backend_config.items() if k != "type"}
-                )
-            elif backend_type.lower() == "claude":
-                agent_config = AgentConfig.create_claude_config(
-                    **{k: v for k, v in backend_config.items() if k != "type"}
-                )
-            elif backend_type.lower() == "grok":
-                agent_config = AgentConfig.create_grok_config(
-                    **{k: v for k, v in backend_config.items() if k != "type"}
-                )
+        # Route system_message to backend-specific system prompt parameter
+        system_msg = agent_data.get("system_message")
+        if system_msg:
+            if backend_type_lower == "claude_code":
+                # For Claude Code, use append_system_prompt to preserve Claude Code capabilities
+                agent_config.backend_params["append_system_prompt"] = system_msg
             else:
-                # Fallback to basic config
-                agent_config = AgentConfig(backend_params=backend_config)
-
-            # Set agent ID and system message
-            agent_config.agent_id = agent_config_data.get("id", f"agent{len(agents)+1}")
-            agent_config.custom_system_instruction = agent_config_data.get(
-                "system_message"
-            )
+                # For other backends, fall back to deprecated custom_system_instruction
+                # TODO: Add backend-specific routing for other backends
+                agent_config.custom_system_instruction = system_msg
 
-            agent = ConfigurableAgent(config=agent_config, backend=backend)
-            agents[agent.agent_id] = agent
+        # Timeout configuration will be applied to orchestrator instead of individual agents
 
-    else:
-        raise ConfigurationError(
-            "Configuration must contain either 'agent' or 'agents' section"
-        )
+        agent = ConfigurableAgent(config=agent_config, backend=backend)
+        agents[agent.config.agent_id] = agent
 
     return agents
 
 
 def create_simple_config(
-    backend_type: str, model: str, system_message: Optional[str] = None
+    backend_type: str,
+    model: str,
+    system_message: Optional[str] = None,
+    base_url: Optional[str] = None,
+    ui_config: Optional[Dict[str, Any]] = None,
 ) -> Dict[str, Any]:
     """Create a simple single-agent configuration."""
-    return {
+    backend_config = {"type": backend_type, "model": model}
+    if base_url:
+        backend_config["base_url"] = base_url
+
+    # Add required workspace configuration for Claude Code backend
+    if backend_type == "claude_code":
+        backend_config["cwd"] = "workspace1"
+
+    # Use provided UI config or default to rich_terminal for CLI usage
+    if ui_config is None:
+        ui_config = {"display_type": "rich_terminal", "logging_enabled": True}
+
+    config = {
         "agent": {
             "id": "agent1",
-            "backend": {"type": backend_type, "model": model},
+            "backend": backend_config,
             "system_message": system_message or "You are a helpful AI assistant.",
         },
-        "ui": {"display_type": "rich_terminal", "logging_enabled": True},
+        "ui": ui_config,
     }
 
+    # Add orchestrator config with .massgen/ structure for Claude Code
+    if backend_type == "claude_code":
+        config["orchestrator"] = {
+            "snapshot_storage": ".massgen/snapshots",
+            "agent_temporary_workspace": ".massgen/temp_workspaces",
+            "session_storage": ".massgen/sessions",
+        }
+
+    return config
+
+
+def validate_context_paths(config: Dict[str, Any]) -> None:
+    """Validate that all context paths in the config exist.
+
+    Context paths can be either files or directories.
+    File-level context paths allow access to specific files without exposing sibling files.
+    Raises ConfigurationError with clear message if any paths don't exist.
+    """
+    orchestrator_cfg = config.get("orchestrator", {})
+    context_paths = orchestrator_cfg.get("context_paths", [])
+
+    missing_paths = []
+
+    for context_path_config in context_paths:
+        if isinstance(context_path_config, dict):
+            path = context_path_config.get("path")
+        else:
+            # Handle string format for backwards compatibility
+            path = context_path_config
+
+        if path:
+            path_obj = Path(path)
+            if not path_obj.exists():
+                missing_paths.append(path)
+
+    if missing_paths:
+        errors = ["Context paths not found:"]
+        for path in missing_paths:
+            errors.append(f"  - {path}")
+        errors.append("\nPlease update your configuration with valid paths.")
+        raise ConfigurationError("\n".join(errors))
+
+
+def relocate_filesystem_paths(config: Dict[str, Any]) -> None:
+    """Relocate filesystem paths (orchestrator paths and agent workspaces) to be under .massgen/ directory.
+
+    Modifies the config in-place to ensure all MassGen state is organized
+    under .massgen/ for clean project structure.
+    """
+    massgen_dir = Path(".massgen")
+
+    # Relocate orchestrator paths
+    orchestrator_cfg = config.get("orchestrator", {})
+    if orchestrator_cfg:
+        path_fields = [
+            "snapshot_storage",
+            "agent_temporary_workspace",
+            "session_storage",
+        ]
+
+        for field in path_fields:
+            if field in orchestrator_cfg:
+                user_path = orchestrator_cfg[field]
+                # If user provided an absolute path or already starts with .massgen/, keep as-is
+                if Path(user_path).is_absolute() or user_path.startswith(".massgen/"):
+                    continue
+                # Otherwise, relocate under .massgen/
+                orchestrator_cfg[field] = str(massgen_dir / user_path)
+
+    # Relocate agent workspaces (cwd fields)
+    agent_entries = [config["agent"]] if "agent" in config else config.get("agents", [])
+    for agent_data in agent_entries:
+        backend_config = agent_data.get("backend", {})
+        if "cwd" in backend_config:
+            user_cwd = backend_config["cwd"]
+            # If user provided an absolute path or already starts with .massgen/, keep as-is
+            if Path(user_cwd).is_absolute() or user_cwd.startswith(".massgen/"):
+                continue
+            # Otherwise, relocate under .massgen/workspaces/
+            backend_config["cwd"] = str(massgen_dir / "workspaces" / user_cwd)
+
+
+def load_previous_turns(session_info: Dict[str, Any], session_storage: str) -> List[Dict[str, Any]]:
+    """
+    Load previous turns from session storage.
+
+    Returns:
+        List of previous turn metadata dicts
+    """
+    session_id = session_info.get("session_id")
+    if not session_id:
+        return []
+
+    session_dir = Path(session_storage) / session_id
+    if not session_dir.exists():
+        return []
+
+    previous_turns = []
+    turn_num = 1
+
+    while True:
+        turn_dir = session_dir / f"turn_{turn_num}"
+        if not turn_dir.exists():
+            break
+
+        metadata_file = turn_dir / "metadata.json"
+        if metadata_file.exists():
+            metadata = json.loads(metadata_file.read_text(encoding="utf-8"))
+            # Use absolute path for workspace
+            workspace_path = (turn_dir / "workspace").resolve()
+            previous_turns.append(
+                {
+                    "turn": turn_num,
+                    "path": str(workspace_path),
+                    "task": metadata.get("task", ""),
+                    "winning_agent": metadata.get("winning_agent", ""),
+                },
+            )
+
+        turn_num += 1
+
+    return previous_turns
+
+
+async def handle_session_persistence(
+    orchestrator,
+    question: str,
+    session_info: Dict[str, Any],
+    session_storage: str,
+) -> tuple[Optional[str], int, Optional[str]]:
+    """
+    Handle session persistence after orchestrator completes.
+
+    Returns:
+        tuple: (session_id, updated_turn_number, normalized_answer)
+    """
+    # Get final result from orchestrator
+    final_result = orchestrator.get_final_result()
+    if not final_result:
+        # No filesystem work to persist
+        return (session_info.get("session_id"), session_info.get("current_turn", 0), None)
+
+    # Initialize or reuse session ID
+    session_id = session_info.get("session_id")
+    if not session_id:
+        session_id = f"session_{datetime.now().strftime('%Y%m%d_%H%M%S')}"
+
+    # Increment turn
+    current_turn = session_info.get("current_turn", 0) + 1
+
+    # Create turn directory
+    session_dir = Path(session_storage) / session_id
+    turn_dir = session_dir / f"turn_{current_turn}"
+    turn_dir.mkdir(parents=True, exist_ok=True)
+
+    # Normalize answer paths
+    final_answer = final_result["final_answer"]
+    workspace_path = final_result.get("workspace_path")
+    turn_workspace_path = (turn_dir / "workspace").resolve()  # Make absolute
+
+    if workspace_path:
+        # Replace workspace paths in answer with absolute path
+        normalized_answer = final_answer.replace(workspace_path, str(turn_workspace_path))
+    else:
+        normalized_answer = final_answer
+
+    # Save normalized answer
+    answer_file = turn_dir / "answer.txt"
+    answer_file.write_text(normalized_answer, encoding="utf-8")
+
+    # Save metadata
+    metadata = {
+        "turn": current_turn,
+        "timestamp": datetime.now().isoformat(),
+        "winning_agent": final_result["winning_agent_id"],
+        "task": question,
+        "session_id": session_id,
+    }
+    metadata_file = turn_dir / "metadata.json"
+    metadata_file.write_text(json.dumps(metadata, indent=2), encoding="utf-8")
+
+    # Create/update session summary for easy viewing
+    session_summary_file = session_dir / "SESSION_SUMMARY.txt"
+    summary_lines = []
+
+    if session_summary_file.exists():
+        summary_lines = session_summary_file.read_text(encoding="utf-8").splitlines()
+    else:
+        summary_lines.append("=" * 80)
+        summary_lines.append(f"Multi-Turn Session: {session_id}")
+        summary_lines.append("=" * 80)
+        summary_lines.append("")
+
+    # Add turn separator and info
+    summary_lines.append("")
+    summary_lines.append("=" * 80)
+    summary_lines.append(f"TURN {current_turn}")
+    summary_lines.append("=" * 80)
+    summary_lines.append(f"Timestamp: {metadata['timestamp']}")
+    summary_lines.append(f"Winning Agent: {metadata['winning_agent']}")
+    summary_lines.append(f"Task: {question}")
+    summary_lines.append(f"Workspace: {turn_workspace_path}")
+    summary_lines.append(f"Answer: See {(turn_dir / 'answer.txt').resolve()}")
+    summary_lines.append("")
+
+    session_summary_file.write_text("\n".join(summary_lines), encoding="utf-8")
+
+    # Copy workspace if it exists
+    if workspace_path and Path(workspace_path).exists():
+        shutil.copytree(workspace_path, turn_workspace_path, dirs_exist_ok=True)
+
+    return (session_id, current_turn, normalized_answer)
+
 
 async def run_question_with_history(
     question: str,
     agents: Dict[str, SingleAgent],
     ui_config: Dict[str, Any],
     history: List[Dict[str, Any]],
-) -> str:
-    """Run MassGen with a question and conversation history."""
+    session_info: Dict[str, Any],
+    **kwargs,
+) -> tuple[str, Optional[str], int]:
+    """Run MassGen with a question and conversation history.
+
+    Returns:
+        tuple: (response_text, session_id, turn_number)
+    """
     # Build messages including history
     messages = history.copy()
     messages.append({"role": "user", "content": question})
 
     # Check if we should use orchestrator for single agents (default: False for backward compatibility)
-    use_orchestrator_for_single = ui_config.get(
-        "use_orchestrator_for_single_agent", True
-    )
+    use_orchestrator_for_single = ui_config.get("use_orchestrator_for_single_agent", True)
 
     if len(agents) == 1 and not use_orchestrator_for_single:
         # Single agent mode with history
@@ -290,7 +812,7 @@ async def run_question_with_history(
         print(f"Agent: {agent.agent_id}", flush=True)
         if history:
             print(f"History: {len(history)//2} previous exchanges", flush=True)
-        print(f"Question: {BRIGHT_WHITE}{question}{RESET}", flush=True)
+        print(f"Question: {question}", flush=True)
         print("\n" + "=" * 60, flush=True)
 
         response_content = ""
@@ -305,25 +827,54 @@ async def run_question_with_history(
                 continue
             elif chunk.type == "error":
                 print(f"\n❌ Error: {chunk.error}", flush=True)
-                return ""
+                return ("", session_info.get("session_id"), session_info.get("current_turn", 0))
 
         print("\n" + "=" * 60, flush=True)
-        return response_content
+        # Single agent mode doesn't use session storage
+        return (response_content, session_info.get("session_id"), session_info.get("current_turn", 0))
 
     else:
         # Multi-agent mode with history
-        orchestrator = Orchestrator(agents=agents)
+        # Create orchestrator config with timeout settings
+        timeout_config = kwargs.get("timeout_config")
+        orchestrator_config = AgentConfig()
+        if timeout_config:
+            orchestrator_config.timeout_config = timeout_config
+
+        # Get orchestrator parameters from config
+        orchestrator_cfg = kwargs.get("orchestrator", {})
+
+        # Get context sharing parameters
+        snapshot_storage = orchestrator_cfg.get("snapshot_storage")
+        agent_temporary_workspace = orchestrator_cfg.get("agent_temporary_workspace")
+        session_storage = orchestrator_cfg.get("session_storage", "sessions")  # Default to "sessions"
+
+        # Get debug/test parameters
+        if orchestrator_cfg.get("skip_coordination_rounds", False):
+            orchestrator_config.skip_coordination_rounds = True
+
+        # Load previous turns from session storage for multi-turn conversations
+        previous_turns = load_previous_turns(session_info, session_storage)
+
+        orchestrator = Orchestrator(
+            agents=agents,
+            config=orchestrator_config,
+            snapshot_storage=snapshot_storage,
+            agent_temporary_workspace=agent_temporary_workspace,
+            previous_turns=previous_turns,
+        )
         # Create a fresh UI instance for each question to ensure clean state
         ui = CoordinationUI(
             display_type=ui_config.get("display_type", "rich_terminal"),
             logging_enabled=ui_config.get("logging_enabled", True),
+            enable_final_presentation=True,  # Required for multi-turn: ensures final answer is saved
         )
 
         print(f"\n🤖 {BRIGHT_CYAN}Multi-Agent Mode{RESET}", flush=True)
         print(f"Agents: {', '.join(agents.keys())}", flush=True)
         if history:
             print(f"History: {len(history)//2} previous exchanges", flush=True)
-        print(f"Question: {BRIGHT_WHITE}{question}{RESET}", flush=True)
+        print(f"Question: {question}", flush=True)
         print("\n" + "=" * 60, flush=True)
 
         # For multi-agent with history, we need to use a different approach
@@ -332,29 +883,30 @@ async def run_question_with_history(
         if history and len(history) > 0:
             # Use coordination UI with conversation context
             # Extract current question from messages
-            current_question = (
-                messages[-1].get("content", question) if messages else question
-            )
+            current_question = messages[-1].get("content", question) if messages else question
 
             # Pass the full message context to the UI coordination
-            response_content = await ui.coordinate_with_context(
-                orchestrator, current_question, messages
-            )
+            response_content = await ui.coordinate_with_context(orchestrator, current_question, messages)
         else:
             # Standard coordination for new conversations
             response_content = await ui.coordinate(orchestrator, question)
 
-        return response_content
+        # Handle session persistence if applicable
+        session_id_to_use, updated_turn, normalized_response = await handle_session_persistence(
+            orchestrator,
+            question,
+            session_info,
+            session_storage,
+        )
+
+        # Return normalized response so conversation history has correct paths
+        return (normalized_response or response_content, session_id_to_use, updated_turn)
 
 
-async def run_single_question(
-    question: str, agents: Dict[str, SingleAgent], ui_config: Dict[str, Any]
-) -> str:
+async def run_single_question(question: str, agents: Dict[str, SingleAgent], ui_config: Dict[str, Any], **kwargs) -> str:
     """Run MassGen with a single question."""
     # Check if we should use orchestrator for single agents (default: False for backward compatibility)
-    use_orchestrator_for_single = ui_config.get(
-        "use_orchestrator_for_single_agent", True
-    )
+    use_orchestrator_for_single = ui_config.get("use_orchestrator_for_single_agent", True)
 
     if len(agents) == 1 and not use_orchestrator_for_single:
         # Single agent mode with existing SimpleDisplay frontend
@@ -362,7 +914,7 @@ async def run_single_question(
 
         print(f"\n🤖 {BRIGHT_CYAN}Single Agent Mode{RESET}", flush=True)
         print(f"Agent: {agent.agent_id}", flush=True)
-        print(f"Question: {BRIGHT_WHITE}{question}{RESET}", flush=True)
+        print(f"Question: {question}", flush=True)
         print("\n" + "=" * 60, flush=True)
 
         messages = [{"role": "user", "content": question}]
@@ -384,67 +936,472 @@ async def run_single_question(
 
     else:
         # Multi-agent mode
-        orchestrator = Orchestrator(agents=agents)
+        # Create orchestrator config with timeout settings
+        timeout_config = kwargs.get("timeout_config")
+        orchestrator_config = AgentConfig()
+        if timeout_config:
+            orchestrator_config.timeout_config = timeout_config
+
+        # Get orchestrator parameters from config
+        orchestrator_cfg = kwargs.get("orchestrator", {})
+
+        # Get context sharing parameters
+        snapshot_storage = orchestrator_cfg.get("snapshot_storage")
+        agent_temporary_workspace = orchestrator_cfg.get("agent_temporary_workspace")
+
+        # Get debug/test parameters
+        if orchestrator_cfg.get("skip_coordination_rounds", False):
+            orchestrator_config.skip_coordination_rounds = True
+
+        orchestrator = Orchestrator(
+            agents=agents,
+            config=orchestrator_config,
+            snapshot_storage=snapshot_storage,
+            agent_temporary_workspace=agent_temporary_workspace,
+        )
         # Create a fresh UI instance for each question to ensure clean state
         ui = CoordinationUI(
             display_type=ui_config.get("display_type", "rich_terminal"),
             logging_enabled=ui_config.get("logging_enabled", True),
+            enable_final_presentation=True,  # Ensures final presentation is generated
         )
 
         print(f"\n🤖 {BRIGHT_CYAN}Multi-Agent Mode{RESET}", flush=True)
         print(f"Agents: {', '.join(agents.keys())}", flush=True)
-        print(f"Question: {BRIGHT_WHITE}{question}{RESET}", flush=True)
+        print(f"Question: {question}", flush=True)
         print("\n" + "=" * 60, flush=True)
 
         final_response = await ui.coordinate(orchestrator, question)
         return final_response
 
 
+def prompt_for_context_paths(original_config: Dict[str, Any], orchestrator_cfg: Dict[str, Any]) -> bool:
+    """Prompt user to add context paths in interactive mode.
+
+    Returns True if config was modified, False otherwise.
+    """
+    # Check if filesystem is enabled (at least one agent has cwd)
+    agent_entries = [original_config["agent"]] if "agent" in original_config else original_config.get("agents", [])
+    has_filesystem = any("cwd" in agent.get("backend", {}) for agent in agent_entries)
+
+    if not has_filesystem:
+        return False
+
+    # Show current context paths
+    existing_paths = orchestrator_cfg.get("context_paths", [])
+    cwd = Path.cwd()
+
+    # Use Rich for better display
+    from rich.console import Console as RichConsole
+    from rich.panel import Panel as RichPanel
+
+    rich_console = RichConsole()
+
+    # Build context paths display
+    context_content = []
+    if existing_paths:
+        for path_config in existing_paths:
+            path = path_config.get("path") if isinstance(path_config, dict) else path_config
+            permission = path_config.get("permission", "read") if isinstance(path_config, dict) else "read"
+            context_content.append(f"  [green]✓[/green] {path} [dim]({permission})[/dim]")
+    else:
+        context_content.append("  [yellow]No context paths configured[/yellow]")
+
+    context_panel = RichPanel(
+        "\n".join(context_content),
+        title="[bold bright_cyan]📂 Context Paths[/bold bright_cyan]",
+        border_style="cyan",
+        padding=(0, 2),
+        width=80,
+    )
+    rich_console.print(context_panel)
+    print()
+
+    # Check if CWD is already in context paths
+    cwd_str = str(cwd)
+    cwd_already_added = any((path_config.get("path") if isinstance(path_config, dict) else path_config) == cwd_str for path_config in existing_paths)
+
+    if not cwd_already_added:
+        # Create prompt panel
+        prompt_content = [
+            "[bold cyan]Add current directory as context path?[/bold cyan]",
+            f"  [yellow]{cwd}[/yellow]",
+            "",
+            "  [dim]Context paths give agents access to your project files.[/dim]",
+            "  [dim]• Read-only during coordination (prevents conflicts)[/dim]",
+            "  [dim]• Write permission for final agent to save results[/dim]",
+            "",
+            "  [dim]Options:[/dim]",
+            "  [green]Y[/green] → Add with write permission (default)",
+            "  [cyan]P[/cyan] → Add with protected paths (e.g., .env, secrets)",
+            "  [yellow]N[/yellow] → Skip",
+            "  [blue]C[/blue] → Add custom path",
+        ]
+        prompt_panel = RichPanel(
+            "\n".join(prompt_content),
+            border_style="cyan",
+            padding=(1, 2),
+            width=80,
+        )
+        rich_console.print(prompt_panel)
+        print()
+        try:
+            response = input(f"   {BRIGHT_CYAN}Your choice [Y/P/N/C]:{RESET} ").strip().lower()
+
+            if response in ["y", "yes", ""]:
+                # Add CWD with write permission
+                if "context_paths" not in orchestrator_cfg:
+                    orchestrator_cfg["context_paths"] = []
+                orchestrator_cfg["context_paths"].append({"path": cwd_str, "permission": "write"})
+                print(f"   {BRIGHT_GREEN}✅ Added: {cwd} (write){RESET}", flush=True)
+                return True
+            elif response in ["p", "protected"]:
+                # Add CWD with write permission and protected paths
+                protected_paths = []
+                print(f"\n   {BRIGHT_CYAN}Enter protected paths (one per line, empty to finish):{RESET}", flush=True)
+                print(f"   {BRIGHT_YELLOW}Tip: Protected paths are relative to {cwd}{RESET}", flush=True)
+                while True:
+                    protected_input = input(f"   {BRIGHT_CYAN}→{RESET} ").strip()
+                    if not protected_input:
+                        break
+                    protected_paths.append(protected_input)
+                    print(f"     {BRIGHT_GREEN}✓ Added: {protected_input}{RESET}", flush=True)
+
+                if "context_paths" not in orchestrator_cfg:
+                    orchestrator_cfg["context_paths"] = []
+
+                context_config = {"path": cwd_str, "permission": "write"}
+                if protected_paths:
+                    context_config["protected_paths"] = protected_paths
+
+                orchestrator_cfg["context_paths"].append(context_config)
+                print(f"\n   {BRIGHT_GREEN}✅ Added: {cwd} (write) with {len(protected_paths)} protected path(s){RESET}", flush=True)
+                return True
+            elif response in ["n", "no"]:
+                # User explicitly declined
+                return False
+            elif response in ["c", "custom"]:
+                # Loop until valid path or user cancels
+                print()
+                while True:
+                    custom_path = input(f"   {BRIGHT_CYAN}Enter path (absolute or relative):{RESET} ").strip()
+                    if not custom_path:
+                        print(f"   {BRIGHT_YELLOW}⚠️  Cancelled{RESET}", flush=True)
+                        return False
+
+                    # Resolve to absolute path
+                    abs_path = str(Path(custom_path).resolve())
+
+                    # Check if path exists
+                    if not Path(abs_path).exists():
+                        print(f"   {BRIGHT_RED}✗ Path does not exist: {abs_path}{RESET}", flush=True)
+                        retry = input(f"   {BRIGHT_CYAN}Try again? [Y/n]:{RESET} ").strip().lower()
+                        if retry in ["n", "no"]:
+                            return False
+                        continue
+
+                    # Valid path (file or directory), ask for permission
+                    permission = input(f"   {BRIGHT_CYAN}Permission [read/write] (default: write):{RESET} ").strip().lower() or "write"
+                    if permission not in ["read", "write"]:
+                        permission = "write"
+
+                    # Ask about protected paths if write permission
+                    protected_paths = []
+                    if permission == "write":
+                        add_protected = input(f"   {BRIGHT_CYAN}Add protected paths? [y/N]:{RESET} ").strip().lower()
+                        if add_protected in ["y", "yes"]:
+                            print(f"   {BRIGHT_CYAN}Enter protected paths (one per line, empty to finish):{RESET}", flush=True)
+                            while True:
+                                protected_input = input(f"   {BRIGHT_CYAN}→{RESET} ").strip()
+                                if not protected_input:
+                                    break
+                                protected_paths.append(protected_input)
+                                print(f"     {BRIGHT_GREEN}✓ Added: {protected_input}{RESET}", flush=True)
+
+                    if "context_paths" not in orchestrator_cfg:
+                        orchestrator_cfg["context_paths"] = []
+
+                    context_config = {"path": abs_path, "permission": permission}
+                    if protected_paths:
+                        context_config["protected_paths"] = protected_paths
+
+                    orchestrator_cfg["context_paths"].append(context_config)
+                    if protected_paths:
+                        print(f"   {BRIGHT_GREEN}✅ Added: {abs_path} ({permission}) with {len(protected_paths)} protected path(s){RESET}", flush=True)
+                    else:
+                        print(f"   {BRIGHT_GREEN}✅ Added: {abs_path} ({permission}){RESET}", flush=True)
+                    return True
+            else:
+                # Invalid response - clarify options
+                print(f"\n   {BRIGHT_RED}✗ Invalid option: '{response}'{RESET}", flush=True)
+                print(f"   {BRIGHT_YELLOW}Please choose: Y (yes), P (protected), N (no), or C (custom){RESET}", flush=True)
+                return False
+        except (KeyboardInterrupt, EOFError):
+            print()  # New line after Ctrl+C
+            return False
+
+    return False
+
+
+def show_available_examples():
+    """Display available example configurations from package."""
+    try:
+        from importlib.resources import files
+
+        configs_root = files("massgen") / "configs"
+
+        print(f"\n{BRIGHT_CYAN}Available Example Configurations{RESET}")
+        print("=" * 60)
+
+        # Organize by category
+        categories = {}
+        for config_file in sorted(configs_root.rglob("*.yaml")):
+            # Get relative path from configs root
+            rel_path = str(config_file).replace(str(configs_root) + "/", "")
+            # Extract category (first directory)
+            parts = rel_path.split("/")
+            category = parts[0] if len(parts) > 1 else "root"
+
+            if category not in categories:
+                categories[category] = []
+
+            # Create a short name for @examples/
+            # Use the path without .yaml extension
+            short_name = rel_path.replace(".yaml", "").replace("/", "_")
+
+            categories[category].append((short_name, rel_path))
+
+        # Display categories
+        for category, configs in sorted(categories.items()):
+            print(f"\n{BRIGHT_YELLOW}{category.title()}:{RESET}")
+            for short_name, rel_path in configs[:10]:  # Limit to avoid overwhelming
+                print(f"  {BRIGHT_GREEN}@examples/{short_name:<40}{RESET} {rel_path}")
+
+            if len(configs) > 10:
+                print(f"  ... and {len(configs) - 10} more")
+
+        print(f"\n{BRIGHT_BLUE}Usage:{RESET}")
+        print('  massgen --config @examples/SHORTNAME "Your question"')
+        print("  massgen --example SHORTNAME > my-config.yaml")
+        print()
+
+    except Exception as e:
+        print(f"Error listing examples: {e}")
+        print("Examples may not be available (development mode?)")
+
+
+def print_example_config(name: str):
+    """Print an example config to stdout.
+
+    Args:
+        name: Name of the example (can include or exclude @examples/ prefix)
+    """
+    try:
+        # Remove @examples/ prefix if present
+        if name.startswith("@examples/"):
+            name = name[10:]
+
+        # Try to resolve the config
+        resolved = resolve_config_path(f"@examples/{name}")
+        if resolved:
+            with open(resolved, "r") as f:
+                print(f.read())
+        else:
+            print(f"Error: Could not find example '{name}'", file=sys.stderr)
+            print("Use --list-examples to see available configs", file=sys.stderr)
+            sys.exit(1)
+
+    except ConfigurationError as e:
+        print(f"Error: {e}", file=sys.stderr)
+        sys.exit(1)
+    except Exception as e:
+        print(f"Error printing example config: {e}", file=sys.stderr)
+        sys.exit(1)
+
+
+def should_run_builder() -> bool:
+    """Check if config builder should run automatically.
+
+    Returns True if:
+    - No default config exists at ~/.config/massgen/config.yaml
+    """
+    default_config = Path.home() / ".config/massgen/config.yaml"
+    return not default_config.exists()
+
+
 def print_help_messages():
-    print(
-        "\n💬 Type your questions below. Use slash commands or press Ctrl+C to stop.",
-        flush=True,
+    """Display help messages using Rich for better formatting."""
+    rich_console = Console()
+
+    help_content = """[dim]💬  Type your questions below
+💡  Use slash commands: [cyan]/help[/cyan], [cyan]/quit[/cyan], [cyan]/reset[/cyan], [cyan]/status[/cyan], [cyan]/config[/cyan]
+⌨️   Press [cyan]Ctrl+C[/cyan] to exit[/dim]"""
+
+    help_panel = Panel(
+        help_content,
+        border_style="dim",
+        padding=(0, 2),
+        width=80,
     )
-    print("💡 Commands: /quit, /exit, /reset, /help", flush=True)
-    print("=" * 60, flush=True)
+    rich_console.print(help_panel)
 
 
 async def run_interactive_mode(
-    agents: Dict[str, SingleAgent], ui_config: Dict[str, Any]
+    agents: Dict[str, SingleAgent],
+    ui_config: Dict[str, Any],
+    original_config: Dict[str, Any] = None,
+    orchestrator_cfg: Dict[str, Any] = None,
+    config_path: Optional[str] = None,
+    **kwargs,
 ):
     """Run MassGen in interactive mode with conversation history."""
-    print(f"\n{BRIGHT_CYAN}🤖 MassGen Interactive Mode{RESET}", flush=True)
-    print("="*60, flush=True)
-    
-    # Display configuration
-    print(f"📋 {BRIGHT_YELLOW}Configuration:{RESET}", flush=True)
-    print(f"   Agents: {len(agents)}", flush=True)
-    for agent_id, agent in agents.items():
-        backend_name = agent.backend.__class__.__name__.replace("Backend", "")
-        print(f"   • {agent_id}: {backend_name}", flush=True)
-
-    use_orchestrator_for_single = ui_config.get(
-        "use_orchestrator_for_single_agent", True
+
+    # Use Rich console for better display
+    rich_console = Console()
+
+    # Clear screen
+    rich_console.clear()
+
+    # ASCII art for interactive multi-agent mode
+    ascii_art = """[bold #4A90E2]
+     ███╗   ███╗ █████╗ ███████╗███████╗ ██████╗ ███████╗███╗   ██╗
+     ████╗ ████║██╔══██╗██╔════╝██╔════╝██╔════╝ ██╔════╝████╗  ██║
+     ██╔████╔██║███████║███████╗███████╗██║  ███╗█████╗  ██╔██╗ ██║
+     ██║╚██╔╝██║██╔══██║╚════██║╚════██║██║   ██║██╔══╝  ██║╚██╗██║
+     ██║ ╚═╝ ██║██║  ██║███████║███████║╚██████╔╝███████╗██║ ╚████║
+     ╚═╝     ╚═╝╚═╝  ╚═╝╚══════╝╚══════╝ ╚═════╝ ╚══════╝╚═╝  ╚═══╝[/bold #4A90E2]
+
+     [dim]     🤖 🤖 🤖  →  💬 collaborate  →  🎯 winner  →  📢 final[/dim]
+"""
+
+    # Wrap ASCII art in a panel
+    ascii_panel = Panel(
+        ascii_art,
+        border_style="bold #4A90E2",
+        padding=(0, 2),
+        width=80,
+    )
+    rich_console.print(ascii_panel)
+    print()
+
+    # Create configuration table
+    config_table = Table(
+        show_header=False,
+        box=None,
+        padding=(0, 2),
+        show_edge=False,
     )
+    config_table.add_column("Label", style="bold cyan", no_wrap=True)
+    config_table.add_column("Value", style="white")
+
+    # Determine mode
+    ui_config.get("use_orchestrator_for_single_agent", True)
     if len(agents) == 1:
-        mode = (
-            "Single Agent (Orchestrator)"
-            if use_orchestrator_for_single
-            else "Single Agent (Direct)"
-        )
+        mode = "Single Agent"
+        mode_icon = "🤖"
+    else:
+        mode = f"Multi-Agent ({len(agents)} agents)"
+        mode_icon = "🤝"
+
+    config_table.add_row(f"{mode_icon} Mode:", f"[bold]{mode}[/bold]")
+
+    # Add agents info
+    if len(agents) <= 3:
+        # Show all agents if 3 or fewer
+        for agent_id, agent in agents.items():
+            # Get model name from config
+            model = agent.config.backend_params.get("model", "unknown")
+            backend_name = agent.backend.__class__.__name__.replace("Backend", "")
+            # Show model with backend in parentheses
+            display = f"{model} [dim]({backend_name})[/dim]"
+            config_table.add_row(f"  ├─ {agent_id}:", display)
     else:
-        mode = "Multi-Agent Coordination"
-    print(f"   Mode: {mode}", flush=True)
-    print(f"   UI: {ui_config.get('display_type', 'rich_terminal')}", flush=True)
+        # Show count and first 2 agents
+        agent_list = list(agents.items())
+        for i, (agent_id, agent) in enumerate(agent_list[:2]):
+            model = agent.config.backend_params.get("model", "unknown")
+            backend_name = agent.backend.__class__.__name__.replace("Backend", "")
+            display = f"{model} [dim]({backend_name})[/dim]"
+            config_table.add_row(f"  ├─ {agent_id}:", display)
+        config_table.add_row("  └─ ...", f"[dim]and {len(agents) - 2} more[/dim]")
+
+    # Create main panel with configuration
+    config_panel = Panel(
+        config_table,
+        title="[bold bright_yellow]⚙️  Session Configuration[/bold bright_yellow]",
+        border_style="yellow",
+        padding=(0, 2),
+        width=80,
+    )
+    rich_console.print(config_panel)
+    print()
+
+    # Prompt for context paths if filesystem is enabled
+    if original_config and orchestrator_cfg:
+        config_modified = prompt_for_context_paths(original_config, orchestrator_cfg)
+        if config_modified:
+            # Recreate agents with updated context paths
+            agents = create_agents_from_config(original_config, orchestrator_cfg)
+            print(f"   {BRIGHT_GREEN}✓ Agents reloaded with updated context paths{RESET}", flush=True)
+            print()
 
     print_help_messages()
 
     # Maintain conversation history
     conversation_history = []
 
+    # Session management for multi-turn filesystem support
+    session_id = None
+    current_turn = 0
+    session_storage = kwargs.get("orchestrator", {}).get("session_storage", "sessions")
+
     try:
         while True:
             try:
+                # Recreate agents with previous turn as read-only context path.
+                # This provides agents with BOTH:
+                # 1. Read-only context path (original turn n-1 results) - for reference/comparison
+                # 2. Writable workspace (copy of turn n-1 results, pre-populated by orchestrator) - for modification
+                # This allows agents to compare "what I changed" vs "what was originally there".
+                # TODO: We may want to avoid full recreation if possible in the future, conditioned on being able to easily reset MCPs.
+                if current_turn > 0 and original_config and orchestrator_cfg:
+                    # Get the most recent turn path (the one just completed)
+                    session_dir = Path(session_storage) / session_id
+                    latest_turn_dir = session_dir / f"turn_{current_turn}"
+                    latest_turn_workspace = latest_turn_dir / "workspace"
+
+                    if latest_turn_workspace.exists():
+                        logger.info(f"[CLI] Recreating agents with turn {current_turn} workspace as read-only context path")
+
+                        # Clean up existing agents' backends and filesystem managers
+                        for agent_id, agent in agents.items():
+                            # Cleanup filesystem manager (Docker containers, etc.)
+                            if hasattr(agent, "backend") and hasattr(agent.backend, "filesystem_manager"):
+                                if agent.backend.filesystem_manager:
+                                    try:
+                                        agent.backend.filesystem_manager.cleanup()
+                                    except Exception as e:
+                                        logger.warning(f"[CLI] Cleanup failed for agent {agent_id}: {e}")
+
+                            # Cleanup backend itself
+                            if hasattr(agent.backend, "__aexit__"):
+                                await agent.backend.__aexit__(None, None, None)
+
+                        # Inject previous turn path as read-only context
+                        modified_config = original_config.copy()
+                        agent_entries = [modified_config["agent"]] if "agent" in modified_config else modified_config.get("agents", [])
+
+                        for agent_data in agent_entries:
+                            backend_config = agent_data.get("backend", {})
+                            if "cwd" in backend_config:  # Only inject if agent has filesystem support
+                                existing_context_paths = backend_config.get("context_paths", [])
+                                new_turn_config = {"path": str(latest_turn_workspace.resolve()), "permission": "read"}
+                                backend_config["context_paths"] = existing_context_paths + [new_turn_config]
+
+                        # Recreate agents from modified config
+                        agents = create_agents_from_config(modified_config, orchestrator_cfg)
+                        logger.info(f"[CLI] Successfully recreated {len(agents)} agents with turn {current_turn} path as read-only context")
+
                 question = input(f"\n{BRIGHT_BLUE}👤 User:{RESET} ").strip()
 
                 # Handle slash commands
@@ -465,9 +1422,7 @@ async def run_interactive_mode(
                         )
                         continue
                     elif command in ["/help", "/h"]:
-                        print(
-                            f"\n{BRIGHT_CYAN}📚 Available Commands:{RESET}", flush=True
-                        )
+                        print(f"\n{BRIGHT_CYAN}📚 Available Commands:{RESET}", flush=True)
                         print("   /quit, /exit, /q     - Exit the program", flush=True)
                         print(
                             "   /reset, /clear       - Clear conversation history",
@@ -477,9 +1432,8 @@ async def run_interactive_mode(
                             "   /help, /h            - Show this help message",
                             flush=True,
                         )
-                        print(
-                            "   /status              - Show current status", flush=True
-                        )
+                        print("   /status              - Show current status", flush=True)
+                        print("   /config              - Open config file in editor", flush=True)
                         continue
                     elif command == "/status":
                         print(f"\n{BRIGHT_CYAN}📊 Current Status:{RESET}", flush=True)
@@ -487,15 +1441,9 @@ async def run_interactive_mode(
                             f"   Agents: {len(agents)} ({', '.join(agents.keys())})",
                             flush=True,
                         )
-                        use_orch_single = ui_config.get(
-                            "use_orchestrator_for_single_agent", True
-                        )
+                        use_orch_single = ui_config.get("use_orchestrator_for_single_agent", True)
                         if len(agents) == 1:
-                            mode_display = (
-                                "Single Agent (Orchestrator)"
-                                if use_orch_single
-                                else "Single Agent (Direct)"
-                            )
+                            mode_display = "Single Agent (Orchestrator)" if use_orch_single else "Single Agent (Direct)"
                         else:
                             mode_display = "Multi-Agent"
                         print(f"   Mode: {mode_display}", flush=True)
@@ -503,6 +1451,28 @@ async def run_interactive_mode(
                             f"   History: {len(conversation_history)//2} exchanges",
                             flush=True,
                         )
+                        if config_path:
+                            print(f"   Config: {config_path}", flush=True)
+                        continue
+                    elif command == "/config":
+                        if config_path:
+                            import platform
+                            import subprocess
+
+                            try:
+                                system = platform.system()
+                                if system == "Darwin":  # macOS
+                                    subprocess.run(["open", config_path])
+                                elif system == "Windows":
+                                    subprocess.run(["start", config_path], shell=True)
+                                else:  # Linux and others
+                                    subprocess.run(["xdg-open", config_path])
+                                print(f"\n📝 Opening config file: {config_path}", flush=True)
+                            except Exception as e:
+                                print(f"\n❌ Error opening config file: {e}", flush=True)
+                                print(f"   Config location: {config_path}", flush=True)
+                        else:
+                            print("\n❌ No config file available (using CLI arguments)", flush=True)
                         continue
                     else:
                         print(f"❓ Unknown command: {command}", flush=True)
@@ -530,16 +1500,40 @@ async def run_interactive_mode(
 
                 print(f"\n🔄 {BRIGHT_YELLOW}Processing...{RESET}", flush=True)
 
-                response = await run_question_with_history(
-                    question, agents, ui_config, conversation_history
+                # Increment turn counter BEFORE processing so logs go to correct turn_N directory
+                next_turn = current_turn + 1
+
+                # Initialize session ID on first turn
+                if session_id is None:
+                    session_id = f"session_{datetime.now().strftime('%Y%m%d_%H%M%S')}"
+
+                # Reconfigure logging for the turn we're about to process
+                setup_logging(debug=_DEBUG_MODE, turn=next_turn)
+                logger.info(f"Starting turn {next_turn}")
+
+                # Pass session state for multi-turn filesystem support
+                session_info = {
+                    "session_id": session_id,
+                    "current_turn": current_turn,  # Pass CURRENT turn (for looking up previous turns)
+                    "session_storage": session_storage,
+                }
+                response, updated_session_id, updated_turn = await run_question_with_history(
+                    question,
+                    agents,
+                    ui_config,
+                    conversation_history,
+                    session_info,
+                    **kwargs,
                 )
 
+                # Update session state after completion
+                session_id = updated_session_id
+                current_turn = updated_turn
+
                 if response:
                     # Add to conversation history
                     conversation_history.append({"role": "user", "content": question})
-                    conversation_history.append(
-                        {"role": "assistant", "content": response}
-                    )
+                    conversation_history.append({"role": "assistant", "content": response})
                     print(f"\n{BRIGHT_GREEN}✅ Complete!{RESET}", flush=True)
                     print(
                         f"{BRIGHT_CYAN}💭 History: {len(conversation_history)//2} exchanges{RESET}",
@@ -560,8 +1554,180 @@ async def run_interactive_mode(
     except KeyboardInterrupt:
         print("\n👋 Goodbye!")
 
-async def main():
-    """Main CLI entry point."""
+
+async def main(args):
+    """Main CLI entry point (async operations only)."""
+    # Check if bare `massgen` with no args - use default config if it exists
+    if not args.backend and not args.model and not args.config:
+        # Use resolve_config_path to check project-level then global config
+        resolved_default = resolve_config_path(None)
+        if resolved_default:
+            # Use discovered config for interactive mode (no question) or single query (with question)
+            args.config = str(resolved_default)
+        else:
+            # No default config - this will be handled by wizard trigger in cli_main()
+            if args.question:
+                # User provided a question but no config exists - this is an error
+                print("❌ Configuration error: No default configuration found.", flush=True)
+                print("Run 'massgen --init' to create one, or use 'massgen --model MODEL \"question\"'", flush=True)
+                sys.exit(1)
+            # No question and no config - wizard will be triggered in cli_main()
+            return
+
+    # Validate arguments (only if we didn't auto-set config above)
+    if not args.backend:
+        if not args.model and not args.config:
+            print("❌ Configuration error: Either --config, --model, or --backend must be specified", flush=True)
+            sys.exit(1)
+
+    try:
+        # Load or create configuration
+        if args.config:
+            # Resolve config path (handles @examples/, paths, ~/.config/massgen/agents/)
+            resolved_path = resolve_config_path(args.config)
+            if resolved_path is None:
+                # This shouldn't happen if we reached here, but handle it
+                raise ConfigurationError("Could not resolve config path")
+            config = load_config_file(str(resolved_path))
+            if args.debug:
+                logger.debug(f"Resolved config path: {resolved_path}")
+                logger.debug(f"Config content: {json.dumps(config, indent=2)}")
+        else:
+            model = args.model
+            if args.backend:
+                backend = args.backend
+            else:
+                backend = get_backend_type_from_model(model=model)
+            if args.system_message:
+                system_message = args.system_message
+            else:
+                system_message = None
+            config = create_simple_config(
+                backend_type=backend,
+                model=model,
+                system_message=system_message,
+                base_url=args.base_url,
+            )
+            if args.debug:
+                logger.debug(f"Created simple config with backend: {backend}, model: {model}")
+                logger.debug(f"Config content: {json.dumps(config, indent=2)}")
+
+        # Validate that all context paths exist before proceeding
+        validate_context_paths(config)
+
+        # Relocate all filesystem paths to .massgen/ directory
+        relocate_filesystem_paths(config)
+
+        # Apply command-line overrides
+        ui_config = config.get("ui", {})
+        if args.no_display:
+            ui_config["display_type"] = "simple"
+        if args.no_logs:
+            ui_config["logging_enabled"] = False
+        if args.debug:
+            ui_config["debug"] = True
+            # Enable logging if debug is on
+            ui_config["logging_enabled"] = True
+            # # Force simple UI in debug mode
+            # ui_config["display_type"] = "simple"
+
+        # Apply timeout overrides from CLI arguments
+        timeout_settings = config.get("timeout_settings", {})
+        if args.orchestrator_timeout is not None:
+            timeout_settings["orchestrator_timeout_seconds"] = args.orchestrator_timeout
+
+        # Update config with timeout settings
+        config["timeout_settings"] = timeout_settings
+
+        # Create agents
+        if args.debug:
+            logger.debug("Creating agents from config...")
+        # Extract orchestrator config for agent setup
+        orchestrator_cfg = config.get("orchestrator", {})
+
+        # Check if any agent has cwd (filesystem support) and validate orchestrator config
+        agent_entries = [config["agent"]] if "agent" in config else config.get("agents", [])
+        has_cwd = any("cwd" in agent.get("backend", {}) for agent in agent_entries)
+
+        if has_cwd:
+            if not orchestrator_cfg:
+                raise ConfigurationError(
+                    "Agents with 'cwd' (filesystem support) require orchestrator configuration.\n"
+                    "Please add an 'orchestrator' section to your config file.\n\n"
+                    "Example (customize paths as needed):\n"
+                    "orchestrator:\n"
+                    '  snapshot_storage: "your_snapshot_dir"\n'
+                    '  agent_temporary_workspace: "your_temp_dir"',
+                )
+
+            # Check for required fields in orchestrator config
+            if "snapshot_storage" not in orchestrator_cfg:
+                raise ConfigurationError(
+                    "Missing 'snapshot_storage' in orchestrator configuration.\n"
+                    "This is required for agents with filesystem support (cwd).\n\n"
+                    "Add to your orchestrator section:\n"
+                    '  snapshot_storage: "your_snapshot_dir"  # Directory for workspace snapshots',
+                )
+
+            if "agent_temporary_workspace" not in orchestrator_cfg:
+                raise ConfigurationError(
+                    "Missing 'agent_temporary_workspace' in orchestrator configuration.\n"
+                    "This is required for agents with filesystem support (cwd).\n\n"
+                    "Add to your orchestrator section:\n"
+                    '  agent_temporary_workspace: "your_temp_dir"  # Directory for temporary agent workspaces',
+                )
+
+        agents = create_agents_from_config(config, orchestrator_cfg)
+
+        if not agents:
+            raise ConfigurationError("No agents configured")
+
+        if args.debug:
+            logger.debug(f"Created {len(agents)} agent(s): {list(agents.keys())}")
+
+        # Create timeout config from settings and put it in kwargs
+        timeout_settings = config.get("timeout_settings", {})
+        timeout_config = TimeoutConfig(**timeout_settings) if timeout_settings else TimeoutConfig()
+
+        kwargs = {"timeout_config": timeout_config}
+
+        # Add orchestrator configuration if present
+        if "orchestrator" in config:
+            kwargs["orchestrator"] = config["orchestrator"]
+
+        # Run mode based on whether question was provided
+        try:
+            if args.question:
+                await run_single_question(args.question, agents, ui_config, **kwargs)
+                # if response:
+                #     print(f"\n{BRIGHT_GREEN}Final Response:{RESET}", flush=True)
+                #     print(f"{response}", flush=True)
+            else:
+                # Pass the config path to interactive mode
+                config_file_path = str(resolved_path) if args.config and resolved_path else None
+                await run_interactive_mode(agents, ui_config, original_config=config, orchestrator_cfg=orchestrator_cfg, config_path=config_file_path, **kwargs)
+        finally:
+            # Cleanup all agents' filesystem managers (including Docker containers)
+            for agent_id, agent in agents.items():
+                if hasattr(agent, "backend") and hasattr(agent.backend, "filesystem_manager"):
+                    if agent.backend.filesystem_manager:
+                        try:
+                            agent.backend.filesystem_manager.cleanup()
+                        except Exception as e:
+                            logger.warning(f"[CLI] Cleanup failed for agent {agent_id}: {e}")
+
+    except ConfigurationError as e:
+        print(f"❌ Configuration error: {e}", flush=True)
+        sys.exit(1)
+    except KeyboardInterrupt:
+        print("\n👋 Goodbye!", flush=True)
+    except Exception as e:
+        print(f"❌ Error: {e}", flush=True)
+        sys.exit(1)
+
+
+def cli_main():
+    """Synchronous wrapper for CLI entry point."""
     parser = argparse.ArgumentParser(
         description="MassGen - Multi-Agent Coordination CLI",
         formatter_class=argparse.RawDescriptionHelpFormatter,
@@ -569,21 +1735,40 @@ async def main():
 Examples:
   # Use configuration file
   python -m massgen.cli --config config.yaml "What is machine learning?"
-  
+
   # Quick single agent setup
   python -m massgen.cli --backend openai --model gpt-4o-mini "Explain quantum computing"
   python -m massgen.cli --backend claude --model claude-sonnet-4-20250514 "Analyze this data"
-  
+
+  # Use ChatCompletion backend with custom base URL
+  python -m massgen.cli --backend chatcompletion --model gpt-oss-120b --base-url https://api.cerebras.ai/v1/chat/completions "What is 2+2?"
+
   # Interactive mode
   python -m massgen.cli --config config.yaml
-  
+
+  # Timeout control examples
+  python -m massgen.cli --config config.yaml --orchestrator-timeout 600 "Complex task"
+
   # Create sample configurations
   python -m massgen.cli --create-samples
 
 Environment Variables:
-  OPENAI_API_KEY      - Required for OpenAI backend
-  XAI_API_KEY         - Required for Grok backend  
-  ANTHROPIC_API_KEY   - Required for Claude backend
+    OPENAI_API_KEY      - Required for OpenAI backend
+    XAI_API_KEY         - Required for Grok backend
+    ANTHROPIC_API_KEY   - Required for Claude backend
+    GOOGLE_API_KEY      - Required for Gemini backend (or GEMINI_API_KEY)
+    ZAI_API_KEY         - Required for ZAI backend
+
+    CEREBRAS_API_KEY    - For Cerebras AI (cerebras.ai)
+    TOGETHER_API_KEY    - For Together AI (together.ai, together.xyz)
+    FIREWORKS_API_KEY   - For Fireworks AI (fireworks.ai)
+    GROQ_API_KEY        - For Groq (groq.com)
+    NEBIUS_API_KEY      - For Nebius AI Studio (studio.nebius.ai)
+    OPENROUTER_API_KEY  - For OpenRouter (openrouter.ai)
+    POE_API_KEY         - For POE (poe.com)
+
+  Note: The chatcompletion backend auto-detects the provider from the base_url
+        and uses the appropriate environment variable for API key.
         """,
     )
 
@@ -596,13 +1781,23 @@ Environment Variables:
 
     # Configuration options
     config_group = parser.add_mutually_exclusive_group()
-    config_group.add_argument(
-        "--config", type=str, help="Path to YAML/JSON configuration file"
-    )
+    config_group.add_argument("--config", type=str, help="Path to YAML/JSON configuration file or @examples/NAME")
     config_group.add_argument(
         "--backend",
         type=str,
-        choices=["openai", "grok", "claude", "gemini"],
+        choices=[
+            "chatcompletion",
+            "claude",
+            "gemini",
+            "grok",
+            "openai",
+            "azure_openai",
+            "claude_code",
+            "zai",
+            "lmstudio",
+            "vllm",
+            "sglang",
+        ],
         help="Backend type for quick setup",
     )
 
@@ -610,77 +1805,167 @@ Environment Variables:
     parser.add_argument(
         "--model",
         type=str,
-        default="gpt-4o-mini",
-        help="Model name for quick setup (default: gpt-4o-mini)",
+        default=None,
+        help="Model name for quick setup",
     )
+    parser.add_argument("--system-message", type=str, help="System message for quick setup")
     parser.add_argument(
-        "--system-message", type=str, help="System message for quick setup"
+        "--base-url",
+        type=str,
+        help="Base URL for API endpoint (e.g., https://api.cerebras.ai/v1/chat/completions)",
     )
 
     # UI options
+    parser.add_argument("--no-display", action="store_true", help="Disable visual coordination display")
+    parser.add_argument("--no-logs", action="store_true", help="Disable logging")
+    parser.add_argument("--debug", action="store_true", help="Enable debug mode with verbose logging")
     parser.add_argument(
-        "--no-display", action="store_true", help="Disable visual coordination display"
+        "--init",
+        action="store_true",
+        help="Launch interactive configuration builder to create config file",
+    )
+    parser.add_argument(
+        "--setup-keys",
+        action="store_true",
+        help="Launch interactive API key setup wizard to configure credentials",
+    )
+    parser.add_argument(
+        "--list-examples",
+        action="store_true",
+        help="List available example configurations from package",
+    )
+    parser.add_argument(
+        "--example",
+        type=str,
+        help="Print example config to stdout (e.g., --example basic_multi)",
+    )
+    parser.add_argument(
+        "--show-schema",
+        action="store_true",
+        help="Display configuration schema and available parameters",
+    )
+    parser.add_argument(
+        "--schema-backend",
+        type=str,
+        help="Show schema for specific backend (use with --show-schema)",
+    )
+    parser.add_argument(
+        "--with-examples",
+        action="store_true",
+        help="Include example configurations in schema display",
+    )
+
+    # Timeout options
+    timeout_group = parser.add_argument_group("timeout settings", "Override timeout settings from config")
+    timeout_group.add_argument(
+        "--orchestrator-timeout",
+        type=int,
+        help="Maximum time for orchestrator coordination in seconds (default: 1800)",
     )
-    parser.add_argument("--no-logs", action="store_true", help="Disable logging")
 
     args = parser.parse_args()
 
-    # Validate arguments
-    if not args.backend:
-        if not args.model and not args.config:
-            parser.error(
-                "If there is not --backend, either --config or --model must be specified"
-            )
+    # Always setup logging (will save INFO to file, console output depends on debug flag)
+    setup_logging(debug=args.debug)
 
-    try:
-        # Load or create configuration
-        if args.config:
-            config = load_config_file(args.config)
-        else:
-            model = args.model
-            if args.backend:
-                backend = args.backend
-            else:
-                backend = get_backend_type_from_model(model=model)
-            if args.system_message:
-                system_message = args.system_message
-            else:
-                system_message = None
-            config = create_simple_config(
-                backend_type=backend, model=model, system_message=system_message
-            )
+    if args.debug:
+        logger.info("Debug mode enabled")
+        logger.debug(f"Command line arguments: {vars(args)}")
 
-        # Apply command-line overrides
-        ui_config = config.get("ui", {})
-        if args.no_display:
-            ui_config["display_type"] = "simple"
-        if args.no_logs:
-            ui_config["logging_enabled"] = False
+    # Handle special commands first
+    if args.list_examples:
+        show_available_examples()
+        return
 
-        # Create agents
-        agents = create_agents_from_config(config)
+    if args.example:
+        print_example_config(args.example)
+        return
 
-        if not agents:
-            raise ConfigurationError("No agents configured")
+    if args.show_schema:
+        from .schema_display import show_schema
 
-        # Run mode based on whether question was provided
-        if args.question:
-            response = await run_single_question(args.question, agents, ui_config)
-            # if response:
-            #     print(f"\n{BRIGHT_GREEN}Final Response:{RESET}", flush=True)
-            #     print(f"{response}", flush=True)
+        show_schema(backend=args.schema_backend, show_examples=args.with_examples)
+        return
+
+    # Launch interactive API key setup if requested
+    if args.setup_keys:
+        from .config_builder import ConfigBuilder
+
+        builder = ConfigBuilder()
+        api_keys = builder.interactive_api_key_setup()
+
+        if any(api_keys.values()):
+            print(f"\n{BRIGHT_GREEN}✅ API key setup complete!{RESET}")
+            print(f"{BRIGHT_CYAN}💡 You can now use MassGen with these providers{RESET}\n")
         else:
-            await run_interactive_mode(agents, ui_config)
+            print(f"\n{BRIGHT_YELLOW}⚠️  No API keys configured{RESET}")
+            print(f"{BRIGHT_CYAN}💡 You can run 'massgen --setup-keys' anytime to set them up{RESET}\n")
+        return
+
+    # Launch interactive config builder if requested
+    if args.init:
+        from .config_builder import ConfigBuilder
+
+        builder = ConfigBuilder()
+        result = builder.run()
+
+        if result and len(result) == 2:
+            filepath, question = result
+            if filepath and question:
+                # Update args to use the newly created config
+                args.config = filepath
+                args.question = question
+            elif filepath:
+                # Config created but user chose not to run
+                print(f"\n✅ Configuration saved to: {filepath}")
+                print(f'Run with: python -m massgen.cli --config {filepath} "Your question"')
+                return
+            else:
+                # User cancelled
+                return
+        else:
+            # Builder returned None (cancelled or error)
+            return
+
+    # First-run detection: auto-trigger builder if no config specified and first run
+    if not args.question and not args.config and not args.model and not args.backend:
+        if should_run_builder():
+            print()
+            print()
+            print(f"{BRIGHT_CYAN}{'=' * 60}{RESET}")
+            print(f"{BRIGHT_CYAN}  👋  Welcome to MassGen!{RESET}")
+            print(f"{BRIGHT_CYAN}{'=' * 60}{RESET}")
+            print()
+            print("  Let's set up your default configuration...")
+            print()
+
+            from .config_builder import ConfigBuilder
+
+            builder = ConfigBuilder(default_mode=True)
+            result = builder.run()
+
+            if result and len(result) == 2:
+                filepath, question = result
+                if filepath:
+                    args.config = filepath
+                    if question:
+                        args.question = question
+                    else:
+                        print("\n✅ Configuration saved! You can now run queries.")
+                        print('Example: massgen "Your question here"')
+                        return
+                else:
+                    return
+            else:
+                return
 
-    except ConfigurationError as e:
-        print(f"❌ Configuration error: {e}", flush=True)
-        sys.exit(1)
+    # Now call the async main with the parsed arguments
+    try:
+        asyncio.run(main(args))
     except KeyboardInterrupt:
-        print("\n👋 Goodbye!", flush=True)
-    except Exception as e:
-        print(f"❌ Error: {e}", flush=True)
-        sys.exit(1)
+        # User pressed Ctrl+C - exit gracefully without traceback
+        pass
 
 
 if __name__ == "__main__":
-    asyncio.run(main())
+    cli_main()