It has no problem handling the biggest folders and repos, has compression modes with live preview for compressing giant repos into the available context window of LLMs, secrets detection & optional redaction to keep things like API keys from leaving your machines, less features vs v1 to keep the interface simple and easy to use. It has one main goal which is to get many files into a LLM preferred format to paste into ChatGPT, Claude, any web interface or other. A common workflow when using AI.

Thousands of people use the first version of AI Code Prep GUI because it has a nice user interface, unlike similar tools. But, being written in Python, was too slow for large repos. People didn't complain but I saw that someone took off a star on its github repo page. I figured.. I bet it was the speed issue! Not a problem most of the time, but annoying on large folders with lots of files. Now, even a folder with 20 repos in it, tens of thousands of files, ready in 3 seconds. I also added some other things to make it better like secrets detect (doesn't slow it down at all) & redact (it will swap out an API key for stars or random characters) and at least 5 compression levels that work well on codebases. So you can pack the full repo into one prompt to paste into something like Google AI Studio (often free, still today) or ChatGPT when the weekly limits have been reached (the web interfaces still work ! ) for bug fixing or any other purpose. Its not limited to code - its common to have this same issue come up with many other things when you need many files into a web chat.

The live preview for compression modes is genuinely useful, you can actually see how your repo gets squeezed into the context window before pasting. Stripping v1 down to focus on one solid workflow was the right call too.

The compression-with-live-preview for fitting giant repos into the context window is the interesting part, do you tune the output format per target model, or keep one LLM-preferred format? We route agents like Claude Code and Codex across different upstreams and see meaningfully different tokenization/formatting sweet spots between them.

AI Code Prep v2

AI Code Prep rewritten from the ground up in Rust for speed

AI Code Prep rewritten from the ground up in Rust for speed