WALTER, the GPT Powered GitHub Bot That Writes Code for You

Last weekend, I combined the capabilities of GitHub and GPT-4 to create WALTER, a user-friendly GitHub bot that writes code based on high-level descriptions.

Today, I'm excited to share WALTER with the world by making the source code freely accessible.

By default, WALTER operates in design mode, allowing you to collaborate with the GitHub bot conversationally, brainstorming and planning the ideal solution within a GitHub Issue. This collaboration involves a mix of freeform text and pseudocode, and is modeled after the junior-senior developer workflow.

Once you're satisfied with the proposed approach, you instruct WALTER to write the actual code and submit a pull request for you to review and merge.

As WALTER is GitHub-based, you can pinpoint specific areas in your codebase for the bot to focus on. If required, it fetches code from GitHub links and incorporates it into the conversation, resulting in more effective ideas and code.

WALTER is bootstrapped, meaning it was designed by GPT-4, and much of the code WALTER runs was also written by WALTER.

Use cases:

• Writing idiomatic code
• Planning & implementing pieces of complex features
• Adding documentation to existing code
• Adding tests and test data
• Creating API bindings and services
• Debugging

Check out WALTER on GitHub: https://github.com/octaviuslabs/walter