Branch Switching, Hot Projects, and Reliability Improvements

This update introduces branch switching, a Hot Projects tab for inspiration, and several reliability enhancements.

Branch Switching for More Control

By default, GPT Engineer commits changes directly to the main branch. Now, you have the flexibility to choose which branch GPT Engineer commits to. You can configure this under Projects Settings.

This feature is for those who prefer a PR-focused workflow or that have CI/CD pipelines that automatically deploy on changes to the main branch. It can also serve those that want to manage experimental work separately, ensuring that only finalized updates make their way to your main branch.

Hot Projects Tab for Inspiration

Looking for inspiration or curious to see what others are building with GPT Engineer? We’ve introduced a Hot Projects tab that showcases trending public projects so that you can see what’s generating buzz.

We’ll continue to improve the algorithm that selects these projects, so expect even better recommendations in the coming weeks. You can check out the current hot projects here by visiting the project overview and selecting the Hot tab.

Improved Network Error Detection

GPT Engineer can now detect and surface network errors. If an issue arises with API requests or external services, you’ll be notified with an error, enabling you to quickly address and resolve them.

Better Context Management for Large Projects

More and more of our users move beyond simple prototypes and build larger, more complex applications with GPT Engineer. When projects get big, language models have a harder time attending to the right context, and in some cases GPT Engineer simply wouldn’t work.

To solve this, we've introduced Retrieval-Augmented Generation (RAG), which identifies and retrieves the relevant code for each request. This enhancement allows you to continue building significantly larger projects without compromising the performance of GPT Engineer.

Enhanced Handling of Laziness

Language models have a tendency to be lazy, using placeholders like “// rest of the code” instead of writing the actual code. To address this, we've implemented automatic detection of these placeholders and now seamlessly fill in the blanks to ensure the code is complete and functional.

We're always happy to hear any feedback you have. To reach us, you can:

• Email feedback@gptengineer.app
• Join our Discord communityI