Easy Agent is an open-source, terminal-native project aiming to fully recreate the Claude Code-style local agentic coding experience from the ground up. Targeted at engineers and researchers, it provides a production-style, extensible system for building and understanding advanced local AI coding agents, prioritizing robust architecture and safety boundaries over isolated demos.

How It Works

The project is structured around a five-layer architecture: the Interaction Layer (terminal UI), Orchestration Layer (multi-turn session flow), Core Agentic Loop (reasoning, tool execution, observation), Tooling Layer (file, shell, search capabilities), and Model Communication Layer (LLM API integration). This separation facilitates incremental evolution and maintainability, allowing distinct components like model I/O, actionable capabilities, and user interface to be developed and updated independently.

Quick Start & Requirements

• Install: npm install
• Prerequisites: Node.js, npm, and access to an Anthropic-compatible LLM.
• Environment Variables: ANTHROPIC_MODEL, ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN are required for model communication.
• Development: Run npm run dev.
• Build/Start: Use npm run build and npm start.

Highlighted Details

• Aims to replicate a full Claude Code-class developer experience locally.
• Features a layered, explicit, and extensible architecture.
• Focuses on building a complete local Agent CLI through incremental development.
• The step/ directory contains tutorial-friendly milestone code for learnability.
• A 30-phase roadmap is in place, with foundational layers (0-12) largely completed.

Maintenance & Community

Easy Agent is currently in active reconstruction and is not accepting external contributions to maintain development stability. The project structure and conventions are expected to change frequently. No community links (e.g., Discord, Slack) are provided.

Licensing & Compatibility

The project is licensed under the MIT License, which generally permits commercial use and integration into closed-source projects.

Limitations & Caveats

This repository represents an open-source rebuild in progress, not a finished end-user product. Many advanced features, such as plan mode, task management, sandboxing, and multi-agent collaboration, are still under active development or have not yet been started. External contributions are intentionally postponed.