The repo-to-text tool that GPT-4 mostly wrote itself

What it does

gpt-repository-loader is a Python utility that walks a Git repository and serializes every file into a single text document, preserving directory structure and contents. The resulting blob is meant to be pasted directly into an LLM prompt for tasks like code review or documentation generation. By the author’s own account, it is largely glue code.

The interesting bit

The project’s real payload is meta: it was built mostly by GPT-4, making it a small case study in AI-assisted tooling rather than just another formatter. That recursive quality—an LLM writing the loader that feeds code back to LLMs—gives it more cultural weight than engineering complexity.

Key highlights

• Flattens a Git repo into one text file while preserving file hierarchy
• Supports an optional preamble file to prefix the output with instructions or context
• Outputs to a configurable path or defaults to output.txt
• Includes a basic unit-test harness
• MIT licensed

Verdict

Handy if you routinely dump whole repositories into LLM prompts; less useful if you need selective ingestion or semantic chunking, which the tool does not provide.