Skip to main content
Glama
by 8b-is
QUANTUM_NATIVE_SPEC.md3.51 kB
# Quantum Native Format Specification ## The Vision Smart Tree's quantum scanner emits quantum format **natively** during tree traversal. No intermediate representation, no post-processing - just pure, efficient quantum output as we walk the filesystem. ## Architecture ``` Filesystem → Quantum Scanner → Quantum Stream → [Decoders] → Other Formats ↓ Direct Output (no buffering) ``` ## Token Architecture ### Token ID Space (u16: 0x0000 - 0xFFFF) ``` 0x0000-0x00FF: Reserved System Tokens 0x0001-0x000F: Node types (dir, file, link, etc.) 0x0010-0x001F: Common permissions 0x0020-0x007F: Common extensions 0x0080-0x00FF: Common directory names 0x00A0-0x00AF: Size ranges 0x00B0-0x00BF: Semantic patterns 0x00C0-0x00FF: Reserved 0x0100-0xFFFF: Dynamic User Tokens - Created on-the-fly for frequently seen patterns - Transmitted in header for decoder sync ``` ### Semantic Tokenization Examples ``` Before: package.json → After: [TOKEN: pkg.manifest] Before: node_modules → After: [TOKEN: pkg.node_modules] Before: 0o755 → After: [TOKEN: perm.default_dir] Before: 1024 bytes → After: [TOKEN: size.small] + [8-bit: 4] (4*256 = 1024) ``` ## Stream Format ### Header Section ``` QUANTUM_NATIVE_V1: TOKENS: <token_id>=<pattern> ... DATA: ``` ### Data Section (Binary) Each entry: `[header_byte][data_fields][name][traversal_code]` ### Header Byte Encoding ``` Bit 7: Tokenized name follows Bit 6: Has extended attributes Bit 5: Is symbolic link Bit 4: Is directory Bit 3: Owner/group differ from parent Bit 2: Time differs from parent Bit 1: Permissions differ from parent Bit 0: Has size field ``` ### Traversal Codes - `0x0B` (VT): Same level - `0x0E` (SO): Go deeper (enter directory) - `0x0F` (SI): Go back (exit directory) - `0x0C` (FF): Summary follows ## Compression Advantages 1. **No Redundancy**: Each piece of information appears exactly once 2. **Delta Encoding**: Only differences from parent context 3. **Semantic Tokens**: Common patterns become single bytes 4. **Direct Streaming**: No memory overhead for large trees 5. **SIMD-Friendly**: Aligned data for vector processing ## Example Encoding ``` Directory: src/main.rs (755, 1234 bytes) ``` Traditional JSON (46 bytes): ```json {"name":"src/main.rs","size":1234,"mode":755} ``` Quantum Native (8 bytes): ``` [Header: 0x11] [Size: 0x00 0xD2 0x04] [Token: 0x82] [Token: 0x91] [Traverse: 0x0B] ``` **Compression ratio: 83%** ## Decoder Architecture Other formats are implemented as decoders from quantum: ```rust trait QuantumDecoder { fn decode_entry(&mut self, quantum_entry: &[u8]) -> Result<()>; } struct JsonDecoder { ... } struct ClassicDecoder { ... } struct HexDecoder { ... } ``` ## Future Optimizations 1. **Huffman Coding**: For non-tokenized strings 2. **Run-Length Encoding**: For similar entries 3. **Dictionary Building**: Dynamic token creation 4. **Parallel Processing**: SIMD operations on token streams 5. **Memory Mapping**: Direct filesystem → quantum mapping ## Implementation Status - [x] Basic quantum scanner - [x] Static token map - [x] Size encoding - [x] Permission deltas - [ ] Dynamic tokenization - [ ] Decoder framework - [ ] SIMD optimization - [ ] Streaming compression ## Philosophy "The best format is no format - just pure, semantic information flowing directly from the filesystem to the consumer. Everything else is just a view into this quantum stream." - The Smart Tree Team

Latest Blog Posts

MCP directory API

We provide all the information about MCP servers via our MCP API.

curl -X GET 'https://glama.ai/api/mcp/v1/servers/8b-is/smart-tree'

If you have feedback or need assistance with the MCP directory API, please join our Discord server