Shrink JSON in Prompts (TOON Encoder/Decoder)

Compress, encode, and decode large JSON payloads into a compact, reversible TOON string to reduce token usage in LLM prompts and tool payloads. Use when you need to shrink repeated JSON in prompts ("compress this JSON", "reduce token usage", "encode JSON"), transmit structured data cheaply, or round-trip JSON reliably (encode/decode/validate). Supports schema (key dictionary) mode for additional compression.

toon-json (TOON v1)

Use the bundled script scripts/toon_json.py to encode JSON into a compact TOON string and decode it back losslessly.

Commands

Encode JSON -> TOON

• Minimal:

python3 scripts/toon_json.py encode < input.json

• With schema (better compression for repeated keys):

python3 scripts/toon_json.py encode --schema < input.json

• Pretty wrapper (includes schema metadata):

python3 scripts/toon_json.py encode --schema --pretty < input.json

Decode TOON -> JSON

python3 scripts/toon_json.py decode < input.toon

Derive a schema (key dictionary) from JSON

python3 scripts/toon_json.py schema < input.json

Operational guidance

• Prefer --schema when:
  • The JSON has many repeated keys (APIs, config blobs, tool outputs)
  • You will send similar objects repeatedly across turns
• Prefer schema-less mode when:
  • One-shot payloads
  • You want maximum human readability

Safety / correctness

• The format is designed to be lossless for JSON types: object, array, string, number, boolean, null.
• Rejects non-JSON numbers (NaN/Inf) and non-string object keys.
• If decoding fails, re-run with --pretty output on encode to inspect the embedded schema.