json-squeeze

CLI tool to minify and compress JSON for LLM context windows.

Features

• Strip whitespace - Basic minification (always enabled)
• Shorten keys - Map long keys to short aliases (--shorten-keys)
• Remove nulls - Strip null values from objects (--remove-nulls)
• Truncate strings - Limit string length (--truncate-strings LENGTH)
• Show stats - Display before/after byte counts and compression ratio (--stats)

Installation

# Recommended: uv
uv tool install json-squeeze

# Or: pipx
pipx install json-squeeze

# Or: pip
pip install json-squeeze

# Or: one-liner
curl -fsSL https://raw.githubusercontent.com/Olafs-World/json-squeeze/main/install.sh | bash

Usage

# Basic minification
json-squeeze input.json -o output.json

# All compression options + stats
json-squeeze data.json -k -n -t 100 -s > small.json

# From stdin
cat large.json | json-squeeze --shorten-keys --stats

# Show key mapping (with --shorten-keys)
json-squeeze input.json -k -m -o output.json

Options

Example

Given input.json:

{
  "very_long_key_name": "This is a very long string that we might want to truncate",
  "another_long_key": null,
  "nested_object": {
    "some_property": "value"
  }
}

Run:

json-squeeze input.json -k -n -t 30 -s -m

Output (to stdout):

{"a":"This is a very long string...","b":{"c":"value"}}

Stderr output:

📝 Key Mapping:
  a → very_long_key_name
  b → nested_object
  c → some_property

📊 Compression Stats:
  Original:  187 bytes
  Squeezed:  58 bytes
  Saved:     129 bytes
  Ratio:     69.0% smaller

Why?

When working with LLMs, you often need to fit large JSON payloads into limited context windows. json-squeeze helps by:

1. Removing unnecessary whitespace (basic minification)
2. Shortening verbose key names (e.g., very_long_key_name → a)
3. Stripping null values (often unnecessary)
4. Truncating long strings (when you don't need full text)

This can dramatically reduce token usage while keeping the data structure intact.

License

MIT