CodeBrew

Introduction

What CodeBrew is and how it fits into your terminal workflow.

CodeBrew is a terminal AI coding agent built on the Anthropic Claude Agent SDK. It can read, explore, edit, and run code on your behalf, with permission modes that keep you in control of exactly what it's allowed to touch.

It supports two interaction styles: a single-shot prompt for quick questions, and an interactive chat session that remembers the whole conversation and lets you switch permission modes on the fly.

The codebrew hello welcome banner, rendered in the terminal with the CodeBrew ASCII wordmark and a row of feature highlights: Build faster, Understand your code, Automate repetitive work, and Collaborate with AI.
codebrew hello — the welcome banner

What it does

  • Answers questions about your files and project
  • Explores a codebase and explains how it works
  • Makes changes: creates files, edits code, runs commands, only when you allow it
  • Holds a live conversation that remembers everything said so far

Safety through modes

The core idea is safety through modes. You decide how much freedom the agent gets: ask for read-only answers, plan to review an approach first, or agent to let it get the work done. You can switch between them mid-conversation with a single /mode command.

CodeBrew is published to npm as @buildwithatul/codebrew. The command you type after installing is codebrew.