Summary

nano-claude provides a highly distilled, ~2,000-line TypeScript reimplementation of Anthropic's Claude Code, targeting developers and researchers who need to understand, modify, or build upon its core agentic capabilities. It serves as a lightweight, incremental alternative to the full Claude Code, significantly simplifying comprehension and customization of LLM agent architectures and tool-use patterns.

How It Works

This project meticulously reconstructs Claude Code's functionality using TypeScript and the Bun runtime, emphasizing a minimal codebase and clear architecture. Its core design revolves around an agentic tool-use loop, progressively building features across distinct versions from a basic streaming REPL to advanced session persistence, smart conversation compaction, and multi-agent orchestration. The approach prioritizes understandability by employing a simple readline interface and abstracting complex LLM interactions.

Quick Start & Requirements

Installation is straightforward: npm install, copy .env.example to .env, and edit the .env file with your Anthropic API key. The project is initiated using npm run dev. Key requirements include Node.js/Bun and an API key for LLM interaction.

Highlighted Details

• Versioned Development: Features are incrementally added across 8 tagged versions (v0-v8), each serving as a milestone to illustrate specific concepts: streaming REPL, agentic tool-use, markdown rendering, safety permissions, session persistence, context window management, multi-level CLAUDE.md, persistent memory, and sub-agent orchestration.
• Core Tools: A curated set of 7 tools is provided for environment interaction: bash (with per-command risk classification), read (with line numbers), write, edit, glob (file pattern search), grep (content search), and agent (sub-agent spawning).
• Flexible CLI & Interaction: Offers both interactive REPL and non-interactive modes, supporting session continuation/resumption, model overrides, and fine-grained control over agent behavior, including a --dangerously-skip-permissions flag for advanced use cases.
• In-Chat Commands: A comprehensive set of slash commands (/help, /cost, /clear, /compact, /model, /sessions, /resume, /remember, /forget, /memory) allows for dynamic management of the agent's state and interaction.

Maintenance & Community

No specific details regarding maintainers, community channels (like Discord/Slack), sponsorships, or roadmap were provided in the README.

Licensing & Compatibility

The project is released under the MIT License, which permits broad use, modification, and distribution, including for commercial purposes, without significant copyleft restrictions.

Limitations & Caveats

As a "lightweight, incremental reimplementation," nano-claude may not replicate all functionalities or the full robustness of the original Claude Code. The presence of a --dangerously-skip-permissions flag indicates potential security considerations when executing tools without explicit user confirmation. The project's versioned structure suggests it is actively under development and may undergo further changes.