An open-source autonomous AI agent implementation inspired by Cursor Agent, built on top of Swarms - the enterprise-grade production-ready multi-agent orchestration framework. This production-grade agent can autonomously plan, execute, and complete complex tasks using a combination of Large Language Model reasoning and tool execution.

Built with Swarms Framework - Leveraging the power of Swarms, the leading open-source framework for building production-ready multi-agent systems. Swarms provides the robust infrastructure, agent orchestration, and enterprise-grade reliability that makes this agent possible.

Open Cursor Agent is a sophisticated AI agent capable of:

• Autonomous Task Planning: Breaking down complex tasks into manageable, sequential subtasks
• Multi-Tool Execution: Leveraging various tools including file operations, command execution, and web search
• Intelligent Reasoning: Using LLM-powered thinking to analyze situations and decide next actions
• State Management: Tracking task progress through well-defined execution states
• Error Handling: Robust error detection and recovery mechanisms

• Python 3.8 or higher
• API key for your chosen LLM provider (e.g., OpenAI)

pip3 install -U open-cursor-agent

WORKSPACE_DIR=""
OPENAI_API_KEY=""
ANTHROPIC_API_KEY=""

from open_cursor.main import OpenCursorAgent

# Initialize the agent
agent = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path=".",
)

# Example task
task_description = """
Create a transformer model in pytorch in a file called transformer.py"
"""

result = agent.run(task_description)

print(result)

You can customize which tools the agent has access to by specifying the allowed_tools parameter:

from open_cursor.main import OpenCursorAgent

# Agent with only search and file reading capabilities
search_agent = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
    allowed_tools=["web_search", "read_file"],
)

# Agent with only code/file operation capabilities
code_agent = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
    allowed_tools=["read_file", "write_file", "list_directory", "search_files"],
)

# Agent with command execution capabilities
command_agent = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
    allowed_tools=["execute_command", "read_file"],
)

# Agent with ALL tools - multiple ways to enable all tools:
# Method 1: Don't specify allowed_tools (default is None)
full_agent = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
)

# Method 2: Explicitly pass "all"
full_agent_explicit = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
    allowed_tools=["all"],
)

# Method 3: Pass empty list
full_agent_empty = OpenCursorAgent(
    model_name="gpt-4o",
    workspace_path="./workspace",
    allowed_tools=[],
)

Available Tools:

• read_file: Read contents of files
• write_file: Write content to files
• search_files: Search for files matching patterns
• list_directory: List directory contents
• execute_command: Execute shell commands
• web_search: Search the web for information
• create_directory: Create new directories
• delete_file: Delete files or directories

Special Values for allowed_tools:

• None (default): All tools are available
• [] (empty list): All tools are available
• ["all"]: All tools are available
• ["tool1", "tool2", ...]: Only specified tools are available

Note: Planning and control tools (create_plan, think, complete_task, subtask_done) are always available regardless of the allowed_tools setting, as they are essential for agent operation.

graph LR
    A[User Task] --> B[Initialize]
    B --> C[Planning]
    C --> D[Execution]
    D --> E[Thinking]
    E --> F{Complete?}
    F -->|No| D
    F -->|Yes| G[Results]
    
    C -.-> H[LLM]
    D -.-> H
    E -.-> H
    D -.-> I[Tools]
    
    style B fill:#4a90e2,color:#fff
    style C fill:#9b59b6,color:#fff
    style D fill:#e74c3c,color:#fff
    style E fill:#f39c12,color:#fff
    style G fill:#27ae60,color:#fff

The agent operates through a state machine with the following phases:

1. Initialization: Task context is created and main task is registered
2. Planning Phase: LLM generates a detailed execution plan with subtasks
3. Execution Phase: Each subtask is executed using appropriate tools
4. Thinking Phase: Results are analyzed and next actions determined
5. Completion: All tasks are finalized and results are returned

• INITIALIZING: Setting up the task context
• PLANNING: Creating a detailed execution plan
• EXECUTING: Performing planned actions
• THINKING: Analyzing results and determining next steps
• COMPLETED: Task successfully finished
• ERROR: Error encountered during execution
• PAUSED: Execution temporarily halted

Contributions are welcome! Please follow these guidelines:

1. Fork the repository
2. Create a feature branch
3. Make your changes with appropriate tests
4. Submit a pull request with a clear description

This project is licensed under the terms specified in the LICENSE file.

Special Thanks: To Swarms Team and the entire Swarms community for building the infrastructure that makes advanced AI agents accessible to everyone. This project stands on the shoulders of giants.