yek

Serialize current directory for LLM consumption

$ yek

$ yek [src/] [tests/]

$ yek --tokens [128k]

$ yek --max-size [10MB]

$ yek | pbcopy

$ yek --line-numbers

$ yek --tree-header

$ yek --json

$ yek "[src/**/*.ts]"

yek [options] [input-paths...]

yek (Farsi for "one") is a fast Rust-based CLI tool that serializes code repositories into text optimized for LLM (Large Language Model) consumption. It combines files into a single output with intelligent ordering and automatic filtering.
The tool respects .gitignore rules, uses Git history to prioritize important files, and automatically skips binary and large files. Output can be split into chunks based on token count or byte size.
When output is piped, yek automatically streams content instead of writing to files. This enables workflows like yek | pbcopy to quickly copy a codebase to clipboard for pasting into an LLM chat.
Configuration can be stored in yek.toml or yek.yaml files for project-specific settings.

--tokens count

Limit output by approximate token count (e.g., 128k, 100)

Limit output by byte size (e.g., 10MB, 128K); default: 10MB

Output results in JSON format

Directory for output files; uses temp directory if not specified

Filename written to current directory

Custom format using FILEPATH and FILECONTENT placeholders

Additional patterns to ignore (extends .gitignore)

Override built-in ignore rules

Include line numbers in output

Include directory tree at the start of output

Show only directory structure without file contents

Skip loading configuration files

Use specific configuration file

Enable debug logging

Token counting is approximate and may vary from actual LLM tokenization. Very large repositories may need chunking with --tokens or --max-size. Glob patterns must be quoted to prevent shell expansion.

yek was created by Mohsen Azimi as a high-performance tool for preparing code for LLM analysis. Written in Rust, it achieved significant speed improvements over similar tools—benchmarks show it running 230× faster than alternatives like Repomix. The name means "one" in Farsi (یک), reflecting its purpose of combining files into one output.

tree(1), find(1), cat(1), rg(1)