Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Fully recreating Claude Code from scratch, so anyone can learn how it works.
| Date | Stars |
|---|---|
| 2026-07-31 | 888 |
| 2026-08-06 | 896 |
Today
+8 stars today
This week
— stars this week
This month
— stars this month
Momentum
32.0
growth rate 0.00%/day
# Easy Agent An open-source, terminal-native project to fully recreate the Claude Code experience from the ground up.  Easy Agent is a long-horizon engineering project focused on rebuilding a complete local agentic coding system in TypeScript and Node.js. The goal is not to publish isolated demos, but to incrementally construct a production-style coding agent with a clean architecture, strong safety boundaries, multi-turn orchestration, local tool execution, and the extensibility required for a full Claude Code-class developer experience. This repository is the open-source implementation track of that effort. Full documentation will be added over time. For now, this README focuses on the project itself: what it aims to become, how it is structured, and where implementation currently stands. > Chinese version: see [README.zh-CN.md](./README.zh-CN.md) ## Vision Easy Agent aims to become a serious open-source recreation of a modern local coding agent system. Core goals: - Fully recreate the Claude Code-style workflow in an open-source codebase - Keep the architecture layered, explicit, and extensible - Prioritize real engineering systems over toy examples - Evolve incrementally toward a complete local Agent CLI - Preserve a stable path toward persistence, compaction, MCP, skills, sandboxing, sub-agents, multi-agent collaboration, multimodal input, plugins, and packaging ## Project Status **Current stage:** Stage 35 — Plugins and Marketplace, next The implementation track has completed through Stage 34, including the CLI, streaming communication, tool execution, terminal UI, session orchestration, context management, MCP, skills, sandboxing, sub-agents, background execution, Agent Teams, hooks, output styles, user commands, rendering upgrades, unified configuration, file history, resilience, headless print mode, classifier-driven Auto Mode, multi-provider streaming, the core-tool expansion for Web, MultiEdit, MCP resources, and PowerShell, multimodal image input, the expanded built-in command set, and Extended Thinking controls and display. Stage 35 is the next planned area: Plugins and Marketplace. The single-file `step/` snapshots now cover Stages 1–34, so the completed implementation chapters can be studied from focused standalone files as well as from the main source tree. Easy Agent should currently be understood as a serious open-source rebuild in progress rather than a finished end-user product. ## Architecture Easy Agent is being built around a five-layer architecture: ```text +---------------------------------------------------+ | 1. Interaction Layer | | Terminal UI, input handling, rendering | +---------------------------------------------------+ | 2. Orchestration Layer | | Multi-turn session flow, usage, commands | +---------------------------------------------------+ | 3. Core Agentic Loop | | Reason -> tool call -> observe -> continue | +---------------------------------------------------+ | 4. Tooling Layer | | File, shell, search, web, MCP, local actions | +---------------------------------------------------+ | 5. Model Communication Layer | | Provider profiles and streaming LLM I/O | +---------------------------------------------------+ ``` This separation makes the system easier to evolve: - the **communication layer** handles provider selection, request translation, and streaming model I/O - the **tool layer** exposes actionable capabilities - the **agentic loop** drives single-turn autonomous execution - the **orchestration layer** manages multi-turn state and control flow - the **interaction layer** turns the runtime into a usable terminal product ## Repository Layout ```text easy-agent/ ├── src/ │ ├── entrypoint/ # CLI bootstrap │ ├── ui/ # React/Ink terminal
Excerpt of 11,595 characters
Read on GitHub
ConardLi · China
70
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:af52aee05c82580b, llm:Repository description: 'Fully recreating Claude Code from scratch, so anyone can learn how it works.' Language: TypeScript. Project name 'easy-agent' suggests an implementation of an AI agent (Claude Code recreation).
matched fp:af52aee05c82580b, llm:Repository description: 'Fully recreating Claude Code from scratch, so anyone can learn how it works.' Language: TypeScript. Project name 'easy-agent' suggests an implementation of an AI agent (Claude Code recreation).
matched fp:af52aee05c82580b, llm:Repository description: 'Fully recreating Claude Code from scratch, so anyone can learn how it works.' Language: TypeScript. Project name 'easy-agent' suggests an implementation of an AI agent (Claude Code recreation).