nanocode

Minimal Claude Code alternative with optional support for OpenAI’s Codex models.

The original version of nanocode was a single Python file (~250 lines) that connected to Anthropic’s Claude models via the /v1/messages API and implemented a simple agentic loop with tool use. This fork extends the project to work with OpenAI models (including Codex) by toggling the MODEL environment variable and providing an appropriate API key. When MODEL names a Claude model (e.g. starts with claude) the script continues to use the Anthropic API【911796291458355†L45-L49】. When it names an OpenAI model (e.g. gpt-5-codex), the script uses the OpenAI Chat Completions or Responses API and authenticates with the OPENAI_API_KEY environment variable【695241305770331†L199-L225】.

Features

• Full agentic loop with tool use
• Tools: read, write, edit, glob, grep, bash
• Conversation history
• Colored terminal output
• Supports both Anthropic and OpenAI models, including Codex

Usage

# Anthropic (default)
export ANTHROPIC_API_KEY="your-anthropic-key"
# Optionally override the model (defaults to claude-opus-4-5)
export MODEL="claude-opus-4-5"
python nanocode.py

# OpenAI / Codex
export OPENAI_API_KEY="your-openai-key"
export MODEL="gpt-5-codex"  # or any supported GPT/Codex model
python nanocode.py

The program inspects the value of MODEL to determine which backend to use. Claude models trigger the Anthropic API, while names beginning with gpt- or containing codex cause the OpenAI API to be used. See the source for details.

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 unless all=true)
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