Split a big repo across files

When the whole repo won't fit in a single message, gnaw can slice the output into several self-contained parts, each kept under a token budget you choose. You paste them one at a time without tripping a model's per-message limit.

The basics

--split-size takes a token budget and needs --output-file as a base name:

gnaw . --split-size 50000 -O context.md

That writes context.part1.md, context.part2.md, and so on — the base name's stem and extension with .partN inserted before the extension. Each part is a complete, independently pasteable document rendered through the normal template, so every part carries its own structure and is valid on its own.

How files are grouped

gnaw walks the selected files in order and fills a part until the next file would push it over the budget, then starts a new part. It's a simple greedy bin-pack by token count, preserving file order.

If a single file is larger than the budget on its own, it can't be split further, so it gets its own part and gnaw prints a warning naming the file and its size. That's expected — lower the budget won't help a lone oversized file; compress it or exclude it instead.

Controlling the parts

Because grouping follows file order and token cost, the two levers that matter are what's included and how big each file is:

Trim first. Apply compression or tighten your --include/--exclude globs so individual files are smaller and pack better.
Find the heavy files. Run with --token-map to see which files dominate, then decide what to compress or drop before splitting.

# Compress, then split into ~40k-token parts
gnaw . --compress moderate --split-size 40000 -O review.md