The rise of AI coding tools like Cursor, Zed, and others has also given rise to code editing tools. One such is the YC backed Codebuff, which allows developers to code with AI directly from their command line interface (CLI).
Earlier known as Manicode, Codebuff has been officially launched to simplify coding tasks by executing natural language requests directly in the terminal. Developed by co-founders Brandon Chen and James Grugett, Codebuff enables developers to edit codebases, run tests, install packages, and more—all without navigating complex menus or UIs.
Users can get started with Codebuff by installing it through npm install -g codebuff and running it from any project directory.
Codebuff’s basic functionality is free to access with up to $20 worth of credits, requiring no sign-ups.
Unlike other code generation tools, Codebuff automatically selects relevant files and performs modifications in response to user input, allowing for a more streamlined workflow. This hands-free approach means developers can simply “chat” with Codebuff to handle project updates, testing, and setup. The tool’s terminal access also lets it execute commands, manage packages, and correct errors, effectively acting as an on-demand junior engineer.
The tool is distinguished by its one-step command system: users type a request, and Codebuff scans the codebase to implement edits directly into the relevant files. It also supports multiple languages and leverages Claude Haiku 3.5 and OpenAI’s models to speed up responses by caching codebase contexts.
Codebuff identifies which files to read or write, saving time and ensuring accurate modifications. The process is entirely command-driven, allowing full project development through simple instructions. Codebuff can run commands directly, installing necessary dependencies and running type checks as needed.
Manicode — the CLI tool that writes code for you — is being renamed to Codebuff!
— James Grugett (@jahooma) November 7, 2024
(Thanks @manicode for being a good sport about us launching with the same name haha!)
We just relaunched on Hacker News and reached #1! Time to celebrate! pic.twitter.com/cWcL8GapgL
Codebuff’s initial challenges, including consistent file editing, were resolved by fine-tuning GPT-4o to handle code modifications accurately—a crucial improvement completed just in time for the Y Combinator interview.
Future plans include an SDK to integrate Codebuff’s capabilities into custom systems, currently available in alpha for early access.
