toon-format/toon

Project overview

TOON is a compact, human-readable encoding of the JSON data model that uses approximately 40% fewer tokens than standard JSON. It is designed for developers and AI engineers who need to reduce LLM input costs and improve parsing reliability for uniform data structures.

Project type

Data Processing

Deployment

Refer to project documentation

License

MIT

Best for

• Developers, AI engineers, and data teams seeking a compact, lossless encoding format to minimize LLM token costs for prompt data.
• Workloads consisting primarily of uniform arrays of objects that benefit from tabular data structures.

Key capabilities

• Provides deterministic, lossless round-trips between JSON and TOON representations.
• Collapses uniform arrays of objects into tables that declare fields once and stream row values line by line.
• Explicit array lengths and field headers give models a clear schema to follow, improving parsing reliability.

Limitations and risks

• For deeply nested or non-uniform structures, JSON-compact often uses fewer tokens than TOON.
• CSV is smaller than TOON for representing pure flat tables.
• The format is stable but the idea is in progress and nothing is set in stone.

Getting started

• You can try TOON instantly with npx. It is installable via npm, making the initial setup difficulty easy.

Evidence and sources

• GitHub project description: 🎒 Token-Oriented Object Notation (TOON) – Compact, human-readable, schema-aware JSON for LLM prompts. Spec, benchmarks, TypeScript SDK.
• README: **Token-Oriented Object Notation** is a compact, human-readable encoding of the JSON data model that minimizes tokens and makes structure easy for models to follow. It's intended…
• README: [](https://www.npmjs.com/package/@toon-format/toon)
• README: AI is becoming cheaper and more accessible, but larger context windows allow for larger data inputs as well. **LLM tokens still cost money** – and standard JSON is verbose and tok…
• README: - 📊 **Token-Efficient & Accurate:** TOON reaches 76.4% accuracy (vs JSON's 75.0%) while using ~40% fewer tokens in mixed-structure benchmarks across 4 models. - 🔁 **JSON Data Mode…

Reviewed at: 2026-07-17T11:06:35Z