massgen 0.1.2__py3-none-any.whl → 0.1.3__py3-none-any.whl

massgen/config_builder.py

@@ -191,27 +191,27 @@ class ConfigBuilder:
         },
         "multimodal": {
             "name": "Multimodal Analysis",
-            "description": "Analyze images, audio, and video content",
+            "description": "Analyze images, audio, video, and documents",
             "recommended_agents": 2,
             "recommended_tools": ["image_understanding", "audio_understanding", "video_understanding"],
             "agent_types": "all",
-            "notes": "Different backends support different modalities",
+            "notes": "Combines custom tools + built-in backend capabilities",
             "info": """[bold cyan]Features auto-configured for this preset:[/bold cyan]
 
-  [green]✓[/green] [bold]Image Understanding[/bold]
-    • Analyze images, screenshots, charts
-    • OCR and text extraction
-    • Available for: OpenAI, Claude Code, Gemini, Azure OpenAI
-
-  [green]✓[/green] [bold]Audio Understanding[/bold] [dim](where supported)[/dim]
-    • Transcribe and analyze audio
-    • Available for: Claude, ChatCompletion
+  [green]✓[/green] [bold]Custom Multimodal Tools (New v0.1.3+)[/bold]
+    • understand_image - Analyze workspace images with gpt-4.1
+    • understand_audio - Transcribe and analyze audio files
+    • understand_video - Extract frames and analyze videos
+    • understand_file - Process documents (PDF, DOCX, XLSX, PPTX)
+    • Works with any backend, processes workspace files
 
-  [green]✓[/green] [bold]Video Understanding[/bold] [dim](where supported)[/dim]
-    • Analyze video content
-    • Available for: Claude, ChatCompletion, OpenAI
+  [green]✓[/green] [bold]Built-in Backend Capabilities[/bold] [dim](passive)[/dim]
+    • Image understanding via upload_files (OpenAI, Claude, Gemini, Azure)
+    • Audio understanding via upload_files (Claude, ChatCompletion)
+    • Video understanding via upload_files (Claude, ChatCompletion, OpenAI)
+    • Image/audio/video generation (where supported)
 
-[dim]Use this for:[/dim] Image analysis, screenshot interpretation, multimedia content analysis.""",
+[dim]Use this for:[/dim] Image analysis, audio transcription, video analysis, document processing.""",
         },
     }
 
@@ -617,6 +617,20 @@ class ConfigBuilder:
             # Build choices for questionary - organized with tool hints
             choices = []
 
+            # Add spacing before first option (using spaces to avoid line)
+            choices.append(questionary.Separator(" "))
+
+            # First option: Browse existing configs (most common for new users)
+            choices.append(
+                questionary.Choice(
+                    title="📦  Browse ready-to-use configs / examples",
+                    value="__browse_existing__",
+                ),
+            )
+            choices.append(questionary.Separator(" "))
+            choices.append(questionary.Separator("┄┄ or build from template ┄┄"))
+            choices.append(questionary.Separator(" "))
+
             # Define display with brief tool descriptions
             display_info = [
                 ("custom", "⚙️", "Custom Configuration", "Choose your own tools"),
@@ -643,12 +657,12 @@ class ConfigBuilder:
                             value=use_case_id,
                         ),
                     )
+
                 except Exception as e:
                     console.print(f"[warning]⚠️  Could not display use case: {e}[/warning]")
 
             # Add helpful context before the prompt
-            console.print("[dim]Choose a preset that matches your task. Each preset auto-configures tools and capabilities.[/dim]")
-            console.print("[dim]You can customize everything in later steps.[/dim]\n")
+            console.print("[dim]Browse ready-to-use configs, or pick a template to build your own.[/dim]\n")
 
             use_case_id = questionary.select(
                 "Select your use case:",
@@ -666,6 +680,10 @@ class ConfigBuilder:
             if use_case_id is None:
                 raise KeyboardInterrupt  # User cancelled, exit immediately
 
+            # Handle special value for browsing existing configs
+            if use_case_id == "__browse_existing__":
+                return "__browse_existing__"
+
             # Show selection with description
             selected_info = self.USE_CASES[use_case_id]
             console.print(f"\n✅ Selected: [green]{selected_info.get('name', use_case_id)}[/green]")
@@ -1521,6 +1539,83 @@ class ConfigBuilder:
 
                         console.print(f"✅ Enabled {len(selected_gen)} generation capability(ies)")
 
+            # Custom multimodal understanding tools (new in v0.1.3+)
+            # Available for ALL use cases - these are active tools that process workspace files
+            console.print()
+            console.print("[cyan]Custom Multimodal Understanding Tools (New in v0.1.3+):[/cyan]")
+            console.print("[dim]These tools let agents analyze workspace files using OpenAI's gpt-4.1 API:[/dim]")
+            console.print("[dim]  • Works with any backend (uses OpenAI for analysis)[/dim]")
+            console.print("[dim]  • Processes files agents generate or discover during execution[/dim]")
+            console.print("[dim]  • Returns structured JSON with detailed metadata[/dim]")
+            console.print("[dim]  • Requires OPENAI_API_KEY in your .env file[/dim]")
+
+            # Default to True for multimodal use case, False for others
+            default_add_mm = use_case == "multimodal"
+
+            if questionary.confirm("Add custom multimodal understanding tools?", default=default_add_mm).ask():
+                # Determine default selections based on use case
+                if use_case == "multimodal":
+                    # For multimodal preset, select all by default
+                    pass
+                elif use_case == "data_analysis":
+                    # For data analysis, suggest image and file tools
+                    pass
+                else:
+                    # For other use cases, none selected by default (let user choose)
+                    pass
+
+                if use_case == "multimodal":
+                    multimodal_tool_choices = [
+                        questionary.Choice("understand_image - Analyze images (PNG, JPEG, JPG)", value="understand_image", checked=True),
+                        questionary.Choice("understand_audio - Transcribe and analyze audio", value="understand_audio", checked=True),
+                        questionary.Choice("understand_video - Extract frames and analyze video", value="understand_video", checked=True),
+                        questionary.Choice("understand_file - Process documents (PDF, DOCX, XLSX, PPTX)", value="understand_file", checked=True),
+                    ]
+                elif use_case == "data_analysis":
+                    multimodal_tool_choices = [
+                        questionary.Choice("understand_image - Analyze images (PNG, JPEG, JPG)", value="understand_image", checked=True),
+                        questionary.Choice("understand_audio - Transcribe and analyze audio", value="understand_audio", checked=False),
+                        questionary.Choice("understand_video - Extract frames and analyze video", value="understand_video", checked=False),
+                        questionary.Choice("understand_file - Process documents (PDF, DOCX, XLSX, PPTX)", value="understand_file", checked=True),
+                    ]
+                else:
+                    multimodal_tool_choices = [
+                        questionary.Choice("understand_image - Analyze images (PNG, JPEG, JPG)", value="understand_image", checked=False),
+                        questionary.Choice("understand_audio - Transcribe and analyze audio", value="understand_audio", checked=False),
+                        questionary.Choice("understand_video - Extract frames and analyze video", value="understand_video", checked=False),
+                        questionary.Choice("understand_file - Process documents (PDF, DOCX, XLSX, PPTX)", value="understand_file", checked=False),
+                    ]
+
+                selected_mm_tools = questionary.checkbox(
+                    "Select custom multimodal tools (Space to select, Enter to confirm):",
+                    choices=multimodal_tool_choices,
+                    style=questionary.Style(
+                        [
+                            ("selected", "fg:cyan"),
+                            ("pointer", "fg:cyan bold"),
+                            ("highlighted", "fg:cyan"),
+                        ],
+                    ),
+                    use_arrow_keys=True,
+                ).ask()
+
+                if selected_mm_tools:
+                    # Initialize custom_tools list if not exists
+                    if "custom_tools" not in agent["backend"]:
+                        agent["backend"]["custom_tools"] = []
+
+                    # Add selected tools
+                    for tool_name in selected_mm_tools:
+                        tool_config = {
+                            "name": [tool_name],
+                            "category": "multimodal",
+                            "path": f"massgen/tool/_multimodal_tools/{tool_name}.py",
+                            "function": [tool_name],
+                        }
+                        agent["backend"]["custom_tools"].append(tool_config)
+
+                    console.print(f"✅ Added {len(selected_mm_tools)} custom multimodal tool(s)")
+
             # MCP servers (custom only)
             # Note: Filesystem is handled internally above, NOT as external MCP
             if "mcp" in provider_info.get("supports", []):
@@ -1807,12 +1902,18 @@ class ConfigBuilder:
                             "code_execution": "💻 Code execution",
                             "web_search": "🔍 Web search",
                             "mcp": "🔌 MCP servers",
+                            "image_understanding": "📷 Image understanding (backend capability)",
+                            "audio_understanding": "🎵 Audio understanding (backend capability)",
+                            "video_understanding": "🎬 Video understanding (backend capability)",
                         }.get(tool, tool)
                         console.print(f"    • {tool_display}")
 
                     if use_case == "coding_docker":
                         console.print("    • 🐳 Docker isolated execution")
 
+                    if use_case == "multimodal":
+                        console.print("    • 🎨 Custom multimodal tools (understand_image, understand_audio, understand_video, understand_file)")
+
                     console.print()
 
                 # Let users select models for each agent
@@ -2302,7 +2403,7 @@ class ConfigBuilder:
                 orchestrator_config = {}
             orchestrator_config["session_storage"] = "sessions"
             console.print()
-            console.print("  ✅ Multi-turn sessions enabled (supports persistent conversations with memory)")
+            console.print("  ✅ Multi-turn sessions enabled (supports persistent conversations)")
 
             # Planning Mode (for MCP irreversible actions) - only ask if MCPs are configured
             has_mcp = any(a.get("backend", {}).get("mcp_servers") for a in agents)
@@ -2321,6 +2422,30 @@ class ConfigBuilder:
                     console.print()
                     console.print("  ✅ Planning mode enabled - MCP tools will plan without executing during coordination")
 
+            # Orchestration Restart Feature
+            console.print()
+            console.print("  [dim]Orchestration Restart: Automatic quality checks with self-correction[/dim]")
+            console.print("  [dim]• Agent evaluates its own answer after coordination[/dim]")
+            console.print("  [dim]• Can restart with specific improvement instructions if incomplete[/dim]")
+            console.print("  [dim]• Each attempt gets isolated logs in attempt_1/, attempt_2/, etc.[/dim]")
+            console.print("  [dim]• Works with all backends (OpenAI, Claude, Gemini, Grok, etc.)[/dim]")
+            console.print("  [dim]• 0 = no restarts (default), 1-2 = recommended, 3 = maximum[/dim]")
+            console.print()
+
+            restart_input = Prompt.ask(
+                "  [prompt]Max orchestration restarts (0-3)[/prompt]",
+                choices=["0", "1", "2", "3"],
+                default="0",
+            )
+
+            max_restarts = int(restart_input)
+            if max_restarts > 0:
+                if "coordination" not in orchestrator_config:
+                    orchestrator_config["coordination"] = {}
+                orchestrator_config["coordination"]["max_orchestration_restarts"] = max_restarts
+                console.print()
+                console.print(f"  ✅ Orchestration restart enabled: up to {max_restarts} restart(s) allowed")
+
             # Voting Sensitivity - only ask for multi-agent setups
             if len(agents) > 1:
                 console.print()
@@ -2634,6 +2759,21 @@ class ConfigBuilder:
                     console.print("[warning]⚠️  No use case selected.[/warning]")
                     return None
 
+                # Handle special case: user wants to browse existing configs
+                if use_case == "__browse_existing__":
+                    console.print("\n[cyan]Opening config selector...[/cyan]\n")
+                    # Import here to avoid circular dependency
+                    from .cli import interactive_config_selector
+
+                    selected_config = interactive_config_selector()
+                    if selected_config:
+                        console.print(f"\n[green]✓ Selected config: {selected_config}[/green]\n")
+                        # Return the selected config as if it was created
+                        return (selected_config, None)
+                    else:
+                        console.print("\n[yellow]⚠️  No config selected[/yellow]\n")
+                        return None
+
                 # Step 2: Configure agents
                 agents = self.configure_agents(use_case, api_keys)
                 if not agents:
@@ -2652,7 +2792,12 @@ class ConfigBuilder:
                 filepath = self.review_and_save(agents, orchestrator_config)
 
                 if filepath:
-                    # Ask if user wants to run now
+                    # In default_mode (first-run), skip "Run now?" and go straight to interactive mode
+                    if self.default_mode:
+                        # Config already saved by review_and_save(), just return to launch interactive mode
+                        return (filepath, None)
+
+                    # In regular --init mode, ask if user wants to run now
                     run_choice = Confirm.ask("\n[prompt]Run MassGen with this configuration now?[/prompt]", default=True)
                     if run_choice is None:
                         raise KeyboardInterrupt  # User cancelled

massgen/configs/README.md

@@ -227,12 +227,61 @@ Most configurations use environment variables for API keys:so
 
 ## Release History & Examples
 
-### v0.1.2 - Latest
+### v0.1.3 - Latest
+**New Features:** Post-Evaluation Workflow, Custom Multimodal Understanding Tools, Docker Sudo Mode
+
+**Configuration Files:**
+- `configs/tools/custom_tools/multimodal_tools/understand_image.yaml` - Image analysis configuration
+- `configs/tools/custom_tools/multimodal_tools/understand_audio.yaml` - Audio transcription configuration
+- `configs/tools/custom_tools/multimodal_tools/understand_video.yaml` - Video analysis configuration
+- `configs/tools/custom_tools/multimodal_tools/understand_file.yaml` - Document processing configuration
+
+**Documentation:**
+- `massgen/tool/docs/multimodal_tools.md` - Complete 779-line multimodal tools guide
+- `docs/source/user_guide/multimodal.rst` - Updated multimodal documentation with custom tools
+- `docs/source/user_guide/code_execution.rst` - Enhanced with 98 lines documenting sudo mode
+- `massgen/docker/README.md` - Updated Docker documentation with sudo mode instructions
+
+**Case Study:**
+- [Multimodal Video Understanding](../../docs/case_studies/multimodal-case-study-video-analysis.md)
+
+**Example Resources:**
+- `configs/resources/v0.1.3-example/multimodality.jpg` - Image example
+- `configs/resources/v0.1.3-example/Sherlock_Holmes.mp3` - Audio example
+- `configs/resources/v0.1.3-example/oppenheimer_trailer_1920.mp4` - Video example
+- `configs/resources/v0.1.3-example/TUMIX.pdf` - PDF document example
+
+**Key Features:**
+- **Post-Evaluation Tools**: Submit and restart capabilities for winning agents with confidence assessments
+- **Multimodal Understanding**: Analyze images, audio, video, and documents using GPT-4.1
+- **Docker Sudo Mode**: Execute privileged commands in containerized environments
+- **Config Builder**: Improved workflow with auto-detection and better provider handling
+
+**Try it:**
+```bash
+# Install or upgrade
+pip install --upgrade massgen
+
+# Try multimodal image understanding
+# (Requires OPENAI_API_KEY in .env)
+massgen --config @examples/tools/custom_tools/multimodal_tools/understand_image \
+  "Please summarize the content in this image."
+
+# Try multimodal audio understanding
+massgen --config @examples/tools/custom_tools/multimodal_tools/understand_audio \
+  "Please summarize the content in this audio."
+
+# Try multimodal video understanding
+massgen --config @examples/tools/custom_tools/multimodal_tools/understand_video \
+  "What's happening in this video?"
+```
+
+### v0.1.2
 **New Features:** Intelligent Planning Mode, Claude 4.5 Haiku Support, Grok Web Search Improvements
 
 **Configuration Files:**
-- `massgen/configs/tools/planning/` - 5 planning mode configurations with selective blocking
-- `massgen/configs/basic/multi/three_agents_default.yaml` - Updated with Grok-4-fast model
+- `configs/tools/planning/` - 5 planning mode configurations with selective blocking
+- `configs/basic/multi/three_agents_default.yaml` - Updated with Grok-4-fast model
 
 **Documentation:**
 - `docs/case_studies/INTELLIGENT_PLANNING_MODE.md` - Complete intelligent planning mode guide
@@ -245,9 +294,6 @@ Most configurations use environment variables for API keys:so
 
 **Try it:**
 ```bash
-# Install or upgrade
-pip install --upgrade massgen
-
 # Try intelligent planning mode with MCP tools
 # (Please read the YAML file for required API keys: DISCORD_TOKEN, OPENAI_API_KEY, etc.)
 massgen --config @examples/tools/planning/five_agents_discord_mcp_planning_mode \

massgen/configs/debug/restart_test_controlled.yaml

@@ -0,0 +1,60 @@
+# Two Agent Orchestration Restart Test Configuration (No Filesystem)
+# This configuration demonstrates the orchestration restart feature for self-correcting coordination
+# Uses debug_final_answer to force restart on first attempt for testing
+#
+# Setup Requirements:
+# 1. Set environment variable in .env file:
+#    OPENAI_API_KEY="your_openai_key"
+#
+# Example usage:
+# massgen --config @examples/configs/debug/restart_test_controlled.yaml "Describe two famous Beatles members: John Lennon and Paul McCartney. Include their birth year, role in the band, and notable contributions."
+#
+# Expected behavior:
+# - Attempt 1: Agents coordinate, final presentation shows debug override (incomplete answer)
+# - Post-evaluation detects incomplete answer and calls restart_orchestration()
+# - Restart banner displays with reason and instructions
+# - Attempt 2: Agents see restart context and produce complete answer
+# - Post-evaluation approves and calls submit()
+# - Inspection menu appears
+
+agents:
+  - id: "agent_a"
+    backend:
+      type: "openai"
+      model: "gpt-5-nano"
+      # type: "gemini"
+      # model: "gemini-2.5-flash"
+      # type: "grok"
+      # model: "grok-4-fast"
+      # type: "claude"
+      # model: "claude-sonnet-4-5-20250929"
+      # type: "claude_code"
+      # model: "claude-sonnet-4-5-20250929"
+      # cwd: "workspace1"
+
+  - id: "agent_b"
+    backend:
+      type: "openai"
+      model: "gpt-5-nano"
+
+orchestrator:
+  snapshot_storage: "snapshots"
+  agent_temporary_workspace: "temp_workspaces"
+
+  # Restart configuration
+  coordination:
+    max_orchestration_restarts: 2  # Allow up to 2 restarts (3 total attempts)
+
+  # DEBUG MODE: Override final answer on attempt 1 to simulate incomplete response
+  debug_final_answer: |
+    John Lennon was born in 1940 in Liverpool, England. He was a founding member
+    of The Beatles and served as rhythm guitarist and vocalist. He co-wrote many
+    of the band's greatest hits with Paul McCartney. After the Beatles disbanded,
+    he pursued a successful solo career and became known for his peace activism.
+
+  # Use lenient voting for faster coordination in debug mode
+  voting_sensitivity: "lenient"
+
+ui:
+  display_type: "rich_terminal"
+  logging_enabled: true

massgen/configs/debug/restart_test_controlled_filesystem.yaml

@@ -0,0 +1,73 @@
+# Two Agent Orchestration Restart Test Configuration with Filesystem
+# This configuration demonstrates the orchestration restart feature with MCP filesystem tools
+# Uses debug_final_answer to force restart on first attempt for testing file creation tasks
+#
+# Setup Requirements:
+# 1. Set environment variable in .env file:
+#    OPENAI_API_KEY="your_openai_key"
+# 2. Filesystem MCP servers will be automatically available via npx
+#
+# Example usage:
+# massgen --config @examples/configs/debug/restart_test_controlled_filesystem.yaml "Create three Python files: main.py with a hello() function, utils.py with a helper() function, and tests.py with a test_hello() function."
+#
+# Expected behavior:
+# - Attempt 1: Agents coordinate with planning mode (no file execution during coordination)
+# - Final presentation shows debug override claiming only main.py was created
+# - Post-evaluation uses MCP tools to verify files and detects missing files
+# - Calls restart_orchestration() with specific instructions
+# - Restart banner displays
+# - Attempt 2: Agents coordinate with restart instructions to create ALL files
+# - Post-evaluation verifies all files exist using MCP tools
+# - Calls submit() - Inspection menu appears
+
+agents:
+  - id: "agent_a"
+    backend:
+      type: "openai"
+      model: "gpt-5-nano"
+      # type: "gemini"
+      # model: "gemini-2.5-flash"
+      # type: "grok"
+      # model: "grok-4-fast"
+      # type: "claude"
+      # model: "claude-sonnet-4-5-20250929"
+      # type: "claude_code"
+      # model: "claude-sonnet-4-5-20250929"
+      cwd: "workspace1"
+
+  # - id: "agent_b"
+  #   backend:
+  #     type: "openai"
+  #     model: "gpt-5-nano"
+  #     cwd: "workspace2"
+
+orchestrator:
+  snapshot_storage: "snapshots"
+  agent_temporary_workspace: "temp_workspaces"
+
+  # Restart configuration
+  coordination:
+    max_orchestration_restarts: 2  # Allow up to 2 restarts (3 total attempts)
+
+  # DEBUG MODE: Override final answer on attempt 1 to simulate incomplete file creation
+  debug_final_answer: |
+    I created a hello.txt file.
+
+   #  file with the following content:
+
+   # ```python
+   # def hello():
+   #     print("Hello, World!")
+
+   # if __name__ == "__main__":
+   #     hello()
+   # ```
+
+  # The file has been created successfully.
+
+  # Use lenient voting for faster coordination in debug mode
+  voting_sensitivity: "lenient"
+
+ui:
+  display_type: "rich_terminal"
+  logging_enabled: true

massgen/configs/tools/code-execution/docker_with_sudo.yaml

@@ -0,0 +1,35 @@
+# Docker Isolation with Sudo Support
+# Run with: uv run python -m massgen.cli --config massgen/configs/tools/code-execution/docker_with_sudo.yaml "Write a factorial function and test it"
+# Expected behavior: Commands execute in isolated Docker container with sudo access for runtime package installation
+#
+# Prerequisites:
+#   1. Docker installed and running
+#   2. Build sudo image FIRST: bash massgen/docker/build.sh --sudo
+#
+# Security: Safe for most use cases - Docker container isolation prevents host access even with sudo.
+
+agent:
+  id: "docker_sudo_agent"
+  backend:
+    type: "openai"
+    model: "gpt-5-mini"
+    cwd: "workspace1"
+
+    # Enable command execution with Docker isolation and sudo
+    enable_mcp_command_line: true
+    command_line_execution_mode: "docker"
+    command_line_docker_enable_sudo: true  # Enables sudo in container
+
+    # Optional: Specify custom image (default auto-switches to sudo variant)
+    # command_line_docker_image: "massgen/mcp-runtime-sudo:latest"
+
+orchestrator:
+  snapshot_storage: "snapshots"
+  agent_temporary_workspace: "temp_workspaces"
+
+timeout_settings:
+  orchestrator_timeout_seconds: 180  # Docker startup adds ~5-10s overhead
+
+ui:
+  display_type: "rich_terminal"
+  logging_enabled: true

massgen/configs/tools/custom_tools/computer_use_browser_example.yaml

@@ -0,0 +1,56 @@
+# MassGen Configuration: Computer Use - Browser Automation
+#
+# This configuration is optimized for browser-based tasks using Playwright.
+#
+# Usage:
+#   massgen --config @examples/tools/custom_tools/computer_use_browser_example "Check the latest OpenAI news on bing.com"
+#
+# Prerequisites:
+#   1. Set OPENAI_API_KEY in your .env file
+#   2. Install Playwright: pip install playwright && playwright install
+
+agents:
+  - id: "browser_automation_agent"
+    backend:
+      type: "openai"
+      model: "computer-use-preview"
+      custom_tools:
+        - name: ["computer_use"]
+          category: "automation"
+          path: "massgen/tool/_computer_use/computer_use_tool.py"
+          function: ["computer_use"]
+          # Default parameters for browser automation
+          default_params:
+            environment: "browser"
+            display_width: 1280
+            display_height: 720
+            max_iterations: 30
+            include_reasoning: true
+
+    system_message: |
+      You are a browser automation specialist with access to the computer_use tool.
+
+      Your capabilities include:
+      - Web navigation and search
+      - Form filling and submission
+      - Data extraction from websites
+      - Multi-step web workflows
+      - Screenshot capture and analysis
+
+      When using the computer_use tool:
+      1. Provide clear, specific task descriptions
+      2. The tool operates in a browser environment (Playwright)
+      3. You'll receive screenshots after each action
+      4. Acknowledge safety warnings when they appear
+      5. Be patient - complex tasks may require multiple iterations
+
+      Best practices:
+      - Break complex tasks into simple steps
+      - Verify each step completed successfully before proceeding
+      - Use wait actions when pages are loading
+      - Be specific about what you're looking for on the page
+
+ui:
+  display_type: "detailed"
+  logging_enabled: true
+  show_screenshots: true

massgen/configs/tools/custom_tools/computer_use_docker_example.yaml

@@ -0,0 +1,65 @@
+# MassGen Configuration: Computer Use - Docker Environment
+#
+# This configuration uses a Docker container for OS-level automation.
+#
+# Usage:
+#   massgen --config @examples/tools/custom_tools/computer_use_docker_example "Open calculator and compute 123 + 456"
+#
+# Prerequisites:
+#   1. Set OPENAI_API_KEY in your .env file
+#   2. Have Docker installed and running
+#   3. Start the Docker container (see README for instructions)
+
+agents:
+  - id: "docker_automation_agent"
+    backend:
+      type: "openai"
+      model: "computer-use-preview"
+      custom_tools:
+        - name: ["computer_use"]
+          category: "automation"
+          path: "massgen/tool/_computer_use/computer_use_tool.py"
+          function: ["computer_use"]
+          # Default parameters for Docker environment
+          default_params:
+            environment: "ubuntu"
+            display_width: 1280
+            display_height: 800
+            max_iterations: 25
+            include_reasoning: true
+            environment_config:
+              container_name: "cua-container"
+              display: ":99"
+
+    system_message: |
+      You are a computer automation specialist with access to the computer_use tool.
+
+      You can control a virtual Ubuntu environment running in Docker to:
+      - Launch and use desktop applications
+      - Perform file operations
+      - Execute system-level tasks
+      - Automate GUI interactions
+
+      The environment includes:
+      - Ubuntu 22.04 desktop (Xfce)
+      - Firefox browser
+      - Standard desktop applications
+      - X11 display system
+
+      When using the computer_use tool:
+      1. Tasks are executed in a sandboxed Docker container
+      2. You have full desktop access via xdotool commands
+      3. Screenshots are captured after each action
+      4. Be specific about coordinates when clicking
+
+      Best practices:
+      - Allow time for applications to launch
+      - Use wait actions between steps
+      - Verify GUI elements are visible before clicking
+      - Consider screen resolution (1280x800)
+
+ui:
+  display_type: "detailed"
+  logging_enabled: true
+  show_screenshots: true
+  show_reasoning: true