Transform your $20 Cursor/Windsurf into a Devin-like experience in one minute! This repository contains configuration files and tools that enhance your Cursor or Windsurf IDE with advanced agentic AI capabilities similar to Devin, including:
- Process planning and self-evolution
- Extended tool usage (web browsing, search, LLM-powered analysis)
- Automated execution (for Windsurf in Docker containers)
- Copy all files from this repository to your project folder
- For Cursor users: The
.cursorrulesfile will be automatically loaded - For Windsurf users: Use both
.windsurfrulesandscratchpad.mdfor similar functionality
This project includes experimental support for a multi-agent system that enhances Cursor's capabilities through a two-agent architecture:
- Planner (powered by OpenAI's o1 model): Handles high-level analysis, task breakdown, and strategic planning
- Executor (powered by Claude): Implements specific tasks, runs tests, and handles implementation details
-
Enhanced Task Quality
- Separation of strategic planning from execution details
- Better cross-checking and validation of solutions
- Iterative refinement through Planner-Executor communication
-
Improved Problem Solving
- Planner can design comprehensive test strategies
- Executor provides detailed feedback and implementation insights
- Continuous communication loop for optimization
A real case study of the multi-agent system debugging the DuckDuckGo search functionality:
-
Initial Analysis
- Planner designed a series of experiments to investigate intermittent search failures
- Executor implemented tests and collected detailed logs
-
Iterative Investigation
- Planner analyzed results and guided investigation to the library's GitHub issues
- Identified a bug in version 6.4 that was fixed in 7.2
-
Solution Implementation
- Planner directed version upgrade and designed comprehensive test cases
- Executor implemented changes and validated with diverse search scenarios
- Final documentation included learnings and cross-checking measures
To use the multi-agent system:
- Switch to the
multi-agentbranch - The system will automatically coordinate between Planner and Executor roles
- Planner uses
tools/plan_exec_llm.pyfor high-level analysis - Executor implements tasks and provides feedback through the scratchpad
This experimental feature transforms the development experience from working with a single assistant to having both a strategic planner and a skilled implementer, significantly improving the depth and quality of task completion.
- Create Python virtual environment:
# Create a virtual environment in ./venv
python3 -m venv venv
# Activate the virtual environment
# On Unix/macOS:
source venv/bin/activate
# On Windows:
.\venv\Scripts\activate- Configure environment variables:
# Copy the example environment file
cp .env.example .env
# Edit .env with your API keys and configurations- Install dependencies:
# Install required packages
pip install -r requirements.txt
# Install Playwright's Chromium browser (required for web scraping)
python -m playwright install chromium- Web scraping with JavaScript support (using Playwright)
- Search engine integration (DuckDuckGo)
- LLM-powered text analysis
- Process planning and self-reflection capabilities
The project includes comprehensive unit tests for all tools. To run the tests:
# Make sure you're in the virtual environment
source venv/bin/activate # On Windows: .\venv\Scripts\activate
# Run all tests
PYTHONPATH=. python -m unittest discover tests/The test suite includes:
- Search engine tests (DuckDuckGo integration)
- Web scraper tests (Playwright-based scraping)
- LLM API tests (OpenAI integration)
For detailed information about the motivation and technical details behind this project, check out the blog post: Turning $20 into $500 - Transforming Cursor into Devin in One Hour
MIT License
