ccline: AI at the zsh prompt level — different approach to the same LLM dev problem

[jianshuo]

Hey rtk community! 👋

I'm a fan of what rtk does — optimizing the LLM interaction layer for dev workflows is exactly the right problem to work on.

I built something that tackles the adjacent problem: when you don't know what command to run at all.

It's called ccline.

What it does

ccline hijacks zsh's command_not_found_handler. If you type something that isn't a real command — plain English, a question, an intention — it sends it to Claude AI and renders Markdown in your terminal. Press Enter to run suggested commands.

$ extract only the http response code from a curl command
→ curl -o /dev/null -s -w "%{http_code}" https://example.com
→ [press Enter to run]

$ recursively delete all .DS_Store files from current directory
→ find . -name ".DS_Store" -delete
→ [press Enter to run]

No prefix, no wrapper command. Just type your intent.

Relationship to rtk's approach

• rtk: reduces token cost for LLM interactions you're already having in your dev tools
• ccline: brings LLM interaction to the place where you 're spending most terminal time — the prompt itself

They're complementary — one optimizes the pipe, one adds a new tap.

Install

curl -fsSL https://raw.githubusercontent.com/jianshuo/ccline/main/install.sh | bash

Requires ANTHROPIC_API_KEY.

→ https://github.com/jianshuo/ccline