AI Coding Agent
for Modern Developers

An open-source AI coding agent that handles web apps, data science pipelines, test generation, and more. Just describe what you need — AACode writes production-ready code with tests, powered by DeepSeek and any LLM of your choice.

Design Principles

🧩

Less Scaffolding

Simple core logic that trusts model capabilities instead of complex frameworks.

📝

File-Based Context

Dynamic discovery with Markdown files as primary storage for context.

Bash Universal Adapter

Flexible system access through intelligent safety guardrails.

🧠

Smart Context Management

Reduction strategies inspired by Cursor and Manus for optimal token usage.

⚙️

Asynchronous Design

All blocking operations are async for maximum performance.

🏗️

Layered Tool System

Atomic tools, management tools, and Skills in a three-layer architecture.

🛡️

Safety Guardrails

Comprehensive command and path security checks built in.

🔌

Extensible Architecture

Support for custom tools, MCP resources, and multiple model backends.

Download Desktop Client

Prefer a graphical interface? Try AACode Desktop.

🍏

macOS

Apple Silicon · v1.7.1

Download .dmg
🪟

Windows

x64 · v1.7.1

Download .exe
ℹ️ The desktop client is a GUI frontend. Python 3.12+ and AACode package (pip install aacode) must be installed separately. See the Desktop Setup Guide below for full instructions.

Quick Start

1

Install AACode

Install via pip, then initialize your configuration.

# Install
pip install aacode

# Initialize configuration
aacode init
2

Set API Key

Configure your LLM provider (supports DeepSeek, OpenAI, and more).

export LLM_API_KEY="your-api-key"
export LLM_API_URL="https://api.deepseek.com/v1"
export LLM_MODEL_NAME="deepseek-chat"
3

Run Your First Task

Point to your project directory and describe what to build.

aacode run -p /path/to/project "Create a calculator program with tests"
4

Desktop Client Setup

Download the desktop app, install Python 3.12+, and start coding visually.

Desktop Setup Guide

1

Install Python 3.12+

AACode Desktop requires Python to run the backend service. Download from python.org or use Homebrew.

# macOS (Homebrew)
brew install python@3.12

# Verify
python3 --version
2

Install AACode

Install the Python package via pip.

pip install aacode
3

Initialize Configuration

Run aacode init or configure API keys in the Desktop settings page.

4

Download & Launch Desktop

Download from the section above, install, and start coding. Select your project directory and describe your task.

Architecture

Core Components of the AACode Engine

🤖

MainAgent

Task decomposition and coordination

🔄

ReActLoop

Intelligent think-act loop

📚

ContextManager

File-based context management

🛠️

Tool System

Atomic, management & Skills layers

🛡️

SafetyGuard

Command & path security checks

⚙️

ConfigManager

Flexible configuration management

MainAgent → ReActLoop → ContextManager + Tools → SafetyGuard

Frequently Asked Questions

What models does AACode support?
AACode supports any OpenAI API-compatible model including DeepSeek, OpenAI GPT-4, Claude (via API gateway), and multimodal models like Kimi K2.5 and MiniMax M2.5. You configure the API endpoint, key, and model name.
Does it work on Windows?
Yes, though it's primarily tested on Linux and macOS. Windows users may encounter Python path issues, but these are easily resolved. The Desktop client provides a smoother Windows experience.
How is AACode different from Cursor or Copilot?
AACode is an open-source CLI-first agent that works with any editor. It focuses on autonomous task execution rather than inline code completion. You describe what you want built, and AACode handles the entire implementation including tests.
Do I need a powerful GPU?
No. AACode is a client-side orchestration tool. All AI processing happens through API calls to LLM providers, so you only need a stable internet connection and an API key.
Can I use my own custom tools or skills?
Yes! AACode has a Skills system. Just create a directory under aacode/skills/ with a SKILL.md description and main.py implementation. It's auto-discovered on restart — no configuration changes needed.
Is it safe to run in my project directory?
AACode includes comprehensive safety features: path security checks to restrict file access within your project, command security verification, AST-based code scanning, and user confirmation for dangerous operations.