1rgs/nanocodePublic

Minimal Claude Code alternative. Single Python file, zero dependencies, ~250 lines.

AI summary: A minimal, ~250 line Python alternative to Claude Code with full agentic loop capabilities.

Stars: 2.5K; +3 today
Forks: 240
Watchers: 14
Open issues: 3
Open PRs: 6
Contributors: ~2
Commits: 11
Branches: 1

PythonNo licenseCreated Jan 11, 2026Last push 6mo ago+13 stars this week+14 this month

Star history

since Jan 11, 2026

2.5K stars as of Aug 6, 2026, tracked back to Jan 11, 2026. Historical curve reconstructed from public GitHub event archives, calibrated to the current total.

Contribution activity

commits per day, last 52 weeks

9 commits in the last yearLessMore

What nanocode does

nanocode is a hyper-minimalist AI coding assistant designed to replicate the core functionality of Claude Code in a single, zero-dependency Python file. Despite being only ~250 lines long, it implements a complete agentic loop capable of autonomous tool use. It connects to the Anthropic API (or OpenRouter) and can execute bash commands, read, write, and edit files, and utilize tools like glob and grep. It serves as both a practical lightweight CLI tool and an educational reference for how to build a capable LLM agent from scratch without bloated frameworks.

nanocode is ideal for Python developers, AI researchers, and minimalists who want to understand agentic loops or need a highly portable CLI coding assistant.

Single File Architecture: The entire application is contained in `nanocode.py`, making it trivial to audit and modify.
Zero Dependencies: Requires no external Python libraries, running purely on the standard library.
Full Agentic Loop: The AI can autonomously decide to use tools, read outputs, and chain actions to solve complex tasks.
Built-in Tools: Natively supports file manipulation (`read`, `write`, `edit`), searching (`grep`, `glob`), and execution (`bash`).
OpenRouter Support: Can easily switch from Anthropic to any model supported by OpenRouter via environment variables.

Where teams use it

Agentic Architecture Study

Developers learning how to build LLM tool-calling loops can read the 250 lines to understand the core concepts.

Instant Deployment

Dropping the single script onto a remote server to have an AI assistant help configure the machine without installing pip packages.

Custom Agent Modification

Forking the single file to quickly add a custom tool (like a database query) for a specific project workflow.

Framework-Free Automation

Using an AI assistant in strict environments where installing large frameworks like LangChain is prohibited.

Getting started: export ANTHROPIC_API_KEY="your-key" && python nanocode.py

README

master branch

nanocode

Minimal Claude Code alternative. Single Python file, zero dependencies, ~250 lines.

Built using Claude Code, then used to build itself.

Features

Full agentic loop with tool use
Tools: read, write, edit, glob, grep, bash
Conversation history
Colored terminal output

Usage

export ANTHROPIC_API_KEY="your-key"
python nanocode.py

OpenRouter

Use OpenRouter to access any model:

export OPENROUTER_API_KEY="your-key"
python nanocode.py

To use a different model:

export OPENROUTER_API_KEY="your-key"
export MODEL="openai/gpt-5.2"
python nanocode.py

Commands

/c - Clear conversation
/q or exit - Quit

Tools

Tool	Description
`read`	Read file with line numbers, offset/limit
`write`	Write content to file
`edit`	Replace string in file (must be unique)
`glob`	Find files by pattern, sorted by mtime
`grep`	Search files for regex
`bash`	Run shell command

Example

────────────────────────────────────────
❯ what files are here?
────────────────────────────────────────

⏺ Glob(**/*.py)
  ⎿  nanocode.py

⏺ There's one Python file: nanocode.py

License

MIT

View on GitHub

Recent activity

commits and pull requests

feat: add request error printClosed
#8 by Grin1024 · updated 3mo ago
fix: typo `pat` to `path`Closed
#12 by vinzloh · updated 4mo ago
Add OpenAI-compatible provider support with NANOCODE_PROVIDER selecti…Open
#15 by greatbody · updated 5mo ago
fix: use context managers and explicit encoding for file operationsOpen
#14 by haosenwang1018 · updated 5mo ago
added adapted scripts. added supports for both gpt and open-weights models. added supported for web search and better markdown rendering.Open
#4 by jaaack-wang · updated 6mo ago
Refactored the code, adding simple tool registration and multi-agent loops Open
#11 by jllike · updated 6mo ago
to completions and ...Closed
#10 by high-u · updated 6mo ago
Add error recovery to prevent tool failure loops. Closed
#9 by josepha-mayo · updated 6mo ago
Merge pull request #2 from charlesrockhead-OR/openrouter-integration
1rgs committed b009d3d · 6mo ago
Merge pull request #3 from charlesrockhead-OR/fix/stream-bash-output
1rgs committed cf1eda5 · 6mo ago
Update README example model to gpt-5.2
charlesrockhead-OR committed 5279d43 · 6mo ago
Rename _OR to OPENROUTER_KEY and add MODEL env var support
charlesrockhead-OR committed 68a29af · 6mo ago

Recent open issues

view all

MIT license used but no LICENSE file
#16 · 3mo ago · by freedevgraph
Model 未调用工具，返回 "I don't have direct access to your file system"
#13 · 6mo ago · by HodenX · 1 comments
Why use Claude when you can use local?
#5 · 6mo ago · by oliverbob · 2 comments

Commits per week

last 52 weeks

9 commits in the last 52 weeks.

When work happens

weekday and hour

Commit volume by weekday and hour (UTC). Larger dots mean more commits.

Who is committing