██▀███ ▄▄▄ ▄▄▄ ██▓▓█████▄
▓██ ▒ ██▒▒████▄ ▒████▄ ▓██▒▒██▀ ██▌
▓██ ░▄█ ▒▒██ ▀█▄ ▒██ ▀█▄ ▒██▒░██ █▌
▒██▀▀█▄ ░██▄▄▄▄██ ░██▄▄▄▄██ ░██░░▓█▄ ▌
░██▓ ▒██▒ ▓█ ▓██▒ ██▓ ▓█ ▓██▒░██░░▒████▓
░ ▒▓ ░▒▓░ ▒▒ ▓▒█░ ▒▓▒ ▒▒ ▓▒█░░▓ ▒▒▓ ▒
░▒ ░ ▒░ ▒ ▒▒ ░ ░▒ ▒ ▒▒ ░ ▒ ░ ░ ▒ ▒
░░ ░ ░ ▒ ░ ░ ▒ ▒ ░ ░ ░ ░
░ ░ ░ ░ ░ ░ ░ ░
░ ░
AI software development agent powered by aider and advanced reasoning models like o1
8000
.
RA.Aid (ReAct Aid) was made by putting aider (https://aider.chat/) in a LangChain ReAct agent loop. This unique combination allows developers to leverage aider's code editing capabilities while benefiting from LangChain's agent-based task execution framework. The tool provides an intelligent assistant that can help with research, planning, and implementation of multi-step development tasks.
RA.Aid is a practical tool for everyday software development and is used for developing real-world applications.
Here's a demo of RA.Aid adding a feature to itself:
- Features
- Installation
- Usage
- Architecture
- Dependencies
- Development Setup
- Contributing
- License
- Contact
👋 Pull requests are very welcome! Have ideas for how to impove RA.Aid? Don't be shy - your help makes a real difference!
💬 Join our Discord community: Click here to join
- This tool can and will automatically execute shell commands and make code changes
- The --cowboy-mode flag can be enabled to skip shell command approval prompts
- No warranty is provided, either express or implied
- Always use in version-controlled repositories
- Review proposed changes in your git diff before committing
-
Multi-Step Task Planning: The agent breaks down complex tasks into discrete, manageable steps and executes them sequentially. This systematic approach ensures thorough implementation and reduces errors.
-
Automated Command Execution: The agent can run shell commands automatically to accomplish tasks. While this makes it powerful, it also means you should carefully review its actions.
-
Ability to Leverage Expert Reasoning Models: The agent can use advanced reasoning models such as OpenAI's o1 just when needed, e.g. to solve complex debugging problems or in planning for complex feature implementation.
-
Web Research Capabilities: Leverages Tavily API for intelligent web searches to enhance research and gather real-world context for development tasks
-
Three-Stage Architecture:
- Research: Analyzes codebases and gathers context
- Planning: Breaks down tasks into specific, actionable steps
- Implementation: Executes each planned step sequentially
What sets RA.Aid apart is its ability to handle complex programming tasks that extend beyond single-shot code edits. By combining research, strategic planning, and implementation into a cohesive workflow, RA.Aid can:
- Break down and execute multi-step programming tasks
- Research and analyze complex codebases to answer architectural questions
- Plan and implement significant code changes across multiple files
- Provide detailed explanations of existing code structure and functionality
- Execute sophisticated refactoring operations with proper planning
-
Three-Stage Architecture: The workflow consists of three powerful stages:
- Research 🔍 - Gather and analyze information
- Planning 📋 - Develop execution strategy
- Implementation ⚡ - Execute the plan with AI assistance
Each stage is powered by dedicated AI agents and specialized toolsets.
-
Advanced AI Integration: Built on LangChain and leverages the latest LLMs for natural language understanding and generation.
-
Human-in-the-Loop Interaction: Optional mode that enables the agent to ask you questions during task execution, ensuring higher accuracy and better handling of complex tasks that may require your input or clarification
-
Comprehensive Toolset:
- Shell command execution
- Expert querying system
- File operations and management
- Memory management
- Research and planning tools
- Code analysis capabilities
-
Interactive CLI Interface: Simple yet powerful command-line interface for seamless interaction
-
Modular Design: Structured as a Python package with specialized modules for console output, processing, text utilities, and tools
-
Git Integration: Built-in support for Git operations and repository management
RA.Aid can be installed directly using pip:
pip install ra-aidBefore using RA.Aid, you'll need:
- Python package
aiderinstalled and available in your PATH:
pip install aider-chat- API keys for the required AI services:
# Set up API keys based on your preferred provider:
# For Anthropic Claude models (recommended)
export ANTHROPIC_API_KEY=your_api_key_here
# For OpenAI models
export OPENAI_API_KEY=your_api_key_here
# For OpenRouter provider (optional)
export OPENROUTER_API_KEY=your_api_key_here
# For OpenAI-compatible providers (optional)
export OPENAI_API_BASE=your_api_base_url
# For Gemini provider (optional)
export GEMINI_API_KEY=your_api_key_here
# For web research capabilities
export TAVILY_API_KEY=your_api_key_hereNote: The programmer tool (aider) will automatically select its model based on your available API keys:
- If ANTHROPIC_API_KEY is set, it will use Claude models
- If only OPENAI_API_KEY is set, it will use OpenAI models
- You can set multiple API keys to enable different features
You can get your API keys from:
- Anthropic API key: https://console.anthropic.com/
- OpenAI API key: https://platform.openai.com/api-keys
- OpenRouter API key: https://openrouter.ai/keys
- Gemini API key: https://aistudio.google.com/app/apikey
RA.Aid is designed to be simple yet powerful. Here's how to use it:
# Basic usage
ra-aid -m "Your task or query here"
# Research-only mode (no implementation)
ra-aid -m "Explain the authentication flow" --research-only
# Enable verbose logging for detailed execution information
ra-aid -m "Add new feature" --verbose-m, --message: The task or query to be executed (required except in chat mode)--research-only: Only perform research without implementation--provider: The LLM provider to use (choices: anthropic, openai, openrouter, openai-compatible, gemini)--model: The model name to use (required for non-Anthropic providers)--research-provider: Provider to use specifically for research tasks (falls back to --provider if not specified)--research-model: Model to use specifically for research tasks (falls back to --model if not specified)--planner-provider: Provider to use specifically for planning tasks (falls back to --provider if not specified)--planner-model: Model to use specifically for planning tasks (falls back to --model if not specified)--cowboy-mode: Skip interactive approval for shell commands--expert-provider: The LLM provider to use for expert knowledge queries (choices: anthropic, openai, openrouter, openai-compatible, gemini)--expert-model: The model name to use for expert knowledge queries (required for non-OpenAI providers)--hil, -H: Enable human-in-the-loop mode for interactive assistance during task execution--chat: Enable chat mode with direct human interaction (implies --hil)--verbose: Enable verbose logging output--temperature: LLM temperature (0.0-2.0) to control randomness in responses--disable-limit-tokens: Disable token limiting for Anthropic Claude react agents--recursion-limit: Maximum recursion depth for agent operations (default: 100)--test-cmd: Custom command to run tests. If set user will be asked if they want to run the test command--auto-test: Automatically run tests after each code change--max-test-cmd-retries: Maximum number of test command retry attempts (default: 3)--version: Show program version number and exit
-
Code Analysis:
ra-aid -m "Explain how the authentication middleware works" --research-only -
Complex Changes:
ra-aid -m "Refactor the database connection code to use connection pooling" --cowboy-mode -
Automated Updates:
ra-aid -m "Update deprecated API calls across the entire codebase" --cowboy-mode -
Code Research:
ra-aid -m "Analyze the current error handling patterns" --research-only -
Code Research:
ra-aid -m "Explain how the authentication middleware works" --research-only -
Refactoring:
ra-aid -m "Refactor the database connection code to use connection pooling" --cowboy-mode
Enable interactive mode to allow the agent to ask you questions during task execution:
ra-aid -m "Implement a new feature" --hil
# or
ra-aid -m "Implement a new feature" -HThis mode is particularly useful for:
- Complex tasks requiring human judgment
- Clarifying ambiguous requirements
- Making architectural decisions
- Validating critical changes
- Providing domain-specific knowledge
The agent features autonomous web research capabilities powered by the Tavily API, seamlessly integrating real-world information into its problem-solving workflow. Web research is conducted automatically when the agent determines additional context would be valuable - no explicit configuration required.
For example, when researching modern authentication practices or investigating new API requirements, the agent will autonomously:
- Search for current best practices and security recommendations
- Find relevant documentation and technical specifications
- Gather real-world implementation examples
- Stay updated on latest industry standards
While web research happens automatically as needed, you can also explicitly request research-focused tasks:
# Focused research task with web search capabilities
ra-aid -m "Research current best practices for API rate limiting" --research-onlyMake sure to set your TAVILY_API_KEY environment variable to enable this feature.
Enable with --chat to transform ra-aid into an interactive assistant that guides you through research and implementation tasks. Have a natural conversation about what you want to build, explore options together, and dispatch work - all while maintaining context of your discussion. Perfect for when you want to think through problems collaboratively rather than just executing commands.
You can interrupt the agent at any time by pressing Ctrl-C. This pauses the agent, allowing you to provide feedback, adjust your instructions, or steer the execution in a new direction. Press Ctrl-C again if you want to completely exit the program.
The --cowboy-mode flag enables automated shell command execution without confirmation prompts. This is useful for:
- CI/CD pipelines
- Automated testing environments
- Batch processing operations
- Scripted workflows
ra-aid -m "Update all deprecated API calls" --cowboy-mode- Cowboy mode skips confirmation prompts for shell commands
- Always use in version-controlled repositories
- Ensure you have a clean working tree before running
- Review changes in git diff before committing
RA.Aid supports multiple AI providers and models. The default model is Anthropic's Claude 3 Sonnet (claude-3-5-sonnet-20241022).
The programmer tool (aider) automatically selects its model based on your available API keys. It will use Claude models if ANTHROPIC_API_KEY is set, or fall back to OpenAI models if only OPENAI_API_KEY is available.
Note: The expert 8000 tool can be configured to use different providers (OpenAI, Anthropic, OpenRouter, Gemini) using the --expert-provider flag along with the corresponding EXPERT_*API_KEY environment variables. Each provider requires its own API key set through the appropriate environment variable.
RA.Aid supports multiple providers through environment variables:
ANTHROPIC_API_KEY: Required for the default Anthropic providerOPENAI_API_KEY: Required for OpenAI providerOPENROUTER_API_KEY: Required for OpenRouter providerDEEPSEEK_API_KEY: Required for DeepSeek providerOPENAI_API_BASE: Required for OpenAI-compatible providers along withOPENAI_API_KEYGEMINI_API_KEY: Required for Gemini provider
Expert Tool Environment Variables:
EXPERT_OPENAI_API_KEY: API key for expert tool using OpenAI providerEXPERT_ANTHROPIC_API_KEY: API key for expert tool using Anthropic providerEXPERT_OPENROUTER_API_KEY: API key for expert tool using OpenRouter providerEXPERT_OPENAI_API_BASE: Base URL for expert tool using OpenAI-compatible providerEXPERT_GEMINI_API_KEY: API key for expert tool using Gemini providerEXPERT_DEEPSEEK_API_KEY: API key for expert tool using DeepSeek provider
You can set these permanently in your shell's configuration file (e.g., ~/.bashrc or ~/.zshrc):
# Default provider (Anthropic)
export ANTHROPIC_API_KEY=your_api_key_here
# For OpenAI features and expert tool
export OPENAI_API_KEY=your_api_key_here
# For OpenRouter provider
export OPENROUTER_API_KEY=your_api_key_here
# For OpenAI-compatible providers
export OPENAI_API_BASE=your_api_base_url
# For Gemini provider
export GEMINI_API_KEY=your_api_key_here-
Using Anthropic (Default)
# Uses default model (claude-3-5-sonnet-20241022) ra-aid -m "Your task" # Or explicitly specify: ra-aid -m "Your task" --provider anthropic --model claude-3-5-sonnet-20241022
-
Using OpenAI
ra-aid -m "Your task" --provider openai --model gpt-4o -
Using OpenRouter
ra-aid -m "Your task" --provider openrouter --model mistralai/mistral-large-2411 -
Using DeepSeek
# Direct DeepSeek provider (requires DEEPSEEK_API_KEY) ra-aid -m "Your task" --provider deepseek --model deepseek-reasoner # DeepSeek via OpenRouter ra-aid -m "Your task" --provider openrouter --model deepseek/deepseek-r1
-
Configuring Expert Provider
The expert tool is used by the agent for complex logic and debugging tasks. It can be configured to use different providers (OpenAI, Anthropic, OpenRouter, Gemini, openai-compatible) using the --expert-provider flag along with the corresponding EXPERT_*API_KEY environment variables.
# Use Anthropic for expert tool export EXPERT_ANTHROPIC_API_KEY=your_anthropic_api_key ra-aid -m "Your task" --expert-provider anthropic --expert-model claude-3-5-sonnet-20241022 # Use OpenRouter for expert tool export OPENROUTER_API_KEY=your_openrouter_api_key ra-aid -m "Your task" --expert-provider openrouter --expert-model mistralai/mistral-large-2411 # Use DeepSeek for expert tool export DEEPSEEK_API_KEY=your_deepseek_api_key ra-aid -m "Your task" --expert-provider deepseek --expert-model deepseek-reasoner # Use default OpenAI for expert tool export EXPERT_OPENAI_API_KEY=your_openai_api_key ra-aid -m "Your task" --expert-provider openai --expert-model o1 # Use Gemini for expert tool export EXPERT_GEMINI_API_KEY=your_gemini_api_key ra-aid -m "Your task" --expert-provider gemini --expert-model gemini-2.0-flash-thinking-exp-1219
Aider specific Environment Variables you can add:
AIDER_FLAGS: Optional comma-separated list of flags to pass to the underlying aider tool (e.g., "yes-always,dark-mode")
# Optional: Configure aider behavior
export AIDER_FLAGS="yes-always,dark-mode,no-auto-commits"Note: For AIDER_FLAGS, you can specify flags with or without the leading --. Multiple flags should be comma-separated, and spaces around flags are automatically handled. For example, both "yes-always,dark-mode" and "--yes-always, --dark-mode" are valid.
Important Notes:
- Performance varies between models. The default Claude 3 Sonnet model currently provides the best and most reliable results.
- Model configuration is done via command line arguments:
--providerand--model - The
--modelargument is required for all providers except Anthropic (which defaults toclaude-3-5-sonnet-20241022)
RA.Aid implements a three-stage architecture for handling development and research tasks:
-
Research Stage:
- Gathers information and context
- Analyzes requirements
- Identifies key components and dependencies
-
Planning Stage:
- Develops detailed implementation plans
- Breaks down tasks into manageable steps
- Identifies potential challenges and solutions
-
Implementation Stage:
- Executes planned tasks
- Generates code or documentation
- Performs necessary system operations
- Console Module (
console/): Handles console output formatting and user interaction - Processing Module (
proc/): Manages interactive processing and workflow control - Text Module (
text/): Provides text processing and manipulation utilities - Tools Module (
tools/): Contains various utility tools for file operations, search, and more
langchain-anthropic: LangChain integration with Anthropic's Claudetavily-python: Tavily API client for web researchlanggraph: Graph-based workflow managementrich>=13.0.0: Terminal formatting and outputGitPython==3.1.41: Git repository managementfuzzywuzzy==0.18.0: Fuzzy string matchingpython-Levenshtein==0.23.0: Fast string matchingpathspec>=0.11.0: Path specification utilities
pytest>=7.0.0: Testing frameworkpytest-timeout>=2.2.0: Test timeout management
- Clone the repository:
git clone https://github.com/ai-christianson/RA.Aid.git
cd RA.Aid- Create and activate a virtual environment:
python -m venv venv
source venv/bin/activate # On Windows use `venv\Scripts\activate`- Install development dependencies:
pip install -r requirements-dev.txt- Run tests:
python -m pytestContributions are welcome! Please follow these steps:
- Fork the repository
- Create a feature branch:
git checkout -b feature/your-feature-name- Make your changes and commit:
git commit -m 'Add some feature'- Push to your fork:
git push origin feature/your-feature-name- Open a Pull Request
- Follow PEP 8 style guidelines
- Add tests for new features
- Update documentation as needed
- Keep commits focused and message clear
- Ensure all tests pass before submitting PR
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.
Copyright (c) 2024 AI Christianson
- Issues: Please report bugs and feature requests on our Issue Tracker
- Repository: https://github.com/ai-christianson/ra-aid
- Documentation: https://github.com/ai-christianson/ra-aid#readme