This repository contains two complementary data formats optimized for AI applications:

Both formats share the JSON data model and provide lossless round-trip encoding.

```bash

npm install @toon-format/toon

npm install @boon-format/boon ```

```ts // TOON - Token-efficient text for LLMs import { encode as toonEncode } from '@toon-format/toon'

const toonOutput = toonEncode({ users: [{ id: 1, name: 'Alice' }] }) // users[1]{id,name}: // 1,Alice

// BOON - Memory-efficient binary for services import { encode as boonEncode, decode as boonDecode } from '@boon-format/boon'

const binaryData = boonEncode({ users: [{ id: 1, name: 'Alice' }] }) // Uint8Array - compact binary representation

const decoded = boonDecode(binaryData) // { users: [{ id: 1, name: 'Alice' }] } ```

Use TOON when:

• Sending data to LLMs (ChatGPT, Claude, etc.)
• Humans need to read or debug the data
• Token costs matter in your AI pipeline
• Working with text-based APIs

Use BOON when:

• Exchanging data between AI services/microservices
• Storing data efficiently in databases or files
• Network bandwidth is a concern
• Working with binary protocols (WebSocket, gRPC)

📖 Full documentation: packages/toon/README.md

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.

JSON is verbose and token-expensive. TOON reduces token count by ~40% while maintaining full JSON compatibility:

JSON (verbose): ```json { "users": [ { "id": 1, "name": "Alice", "role": "admin" }, { "id": 2, "name": "Bob", "role": "user" } ] } ```

TOON (compact): ``` users[2]{id,name,role}: 1,Alice,admin 2,Bob,user ```

• 📊 Token-Efficient: ~40% fewer tokens than JSON
• 🔁 JSON Compatible: Lossless round-trip encoding
• 🛤️ LLM-Friendly: Explicit `[N]` lengths and `{fields}` headers
• 📐 Minimal Syntax: YAML-like readability with CSV-style compactness

```bash npm install @toon-format/toon ```

```ts import { encode, decode } from '@toon-format/toon'

// Encode JSON to TOON const toon = encode({ users: [{ id: 1, name: 'Alice' }] })

// Decode TOON to JSON const json = decode(toon) ```

See the TOON README for complete documentation including benchmarks, CLI usage, and LLM integration guides.

📖 Full documentation: packages/boon/README.md

Binary Object Notation (BOON) is a memory-efficient, cross-platform binary encoding of the JSON data model, designed specifically for AI applications.

BOON provides significant size reductions compared to JSON:

For arrays of objects (common in AI applications), BOON typically achieves 30-60% size reduction compared to JSON.

• 🚀 Memory Efficient: Compact binary encoding with optimized type tags
• 🌐 Cross-Platform: Consistent byte representation using big-endian encoding
• 🤖 AI-Optimized: Designed for efficient data exchange between AI systems
• 🔄 JSON Compatible: Lossless round-trip encoding
• 📡 Streaming Support: Stream events for memory-efficient processing

```bash npm install @boon-format/boon ```

```ts import { encode, decode, decodeStreamSync } from '@boon-format/boon'

// Encode to binary const boon = encode({ users: [{ id: 1, name: 'Alice' }] }) // Returns Uint8Array

// Decode from binary const data = decode(boon)

// Stream decode for large datasets for (const event of decodeStreamSync(boon)) { console.log(event) // { type: 'startObject', keyCount: 1 } // { type: 'key', key: 'users' } // ... } ```

BOON uses a 5-byte header (`BOON` magic + version) followed by type-tagged values:

See the BOON README for complete API documentation.

• Logo design by 鈴木ックス(SZKX)

MIT License © 2025-PRESENT Johann Schopplich