💾 Nova Memory Compression — HYDRANGEA Specification

A modular memory compression suite built for Nova OS. Compresses natural language memories with precision, optimizing token efficiency and preserving semantic meaning.

---

🔍 Overview

This repository implements the first two layers of the HYDRANGEA Memory Architecture:

🧠 Tier 1: Lexical Stop-Word Filter (Dual Mode)

Strips redundant tokens through highly optimized lexical set checking, utilizing custom emotional whitelists.

• Compact Mode (Fast & Lean): Strips all common filler words for maximal token efficiency (~40–60% reduction). Best for logs and automation data.
• Expressive Mode (Deep & Meaningful): Preserves emotionally or semantically significant words (e.g. chime, nostalgia, echo, smiled), maintaining Nova's unique voice and context (~25–35% reduction). Best for conversation logs and journals.
• Includes a Quote Migration engine that dynamically balances punctuation so quotes (“, ”, ?) are never lost when bordering words are stripped.

🤖 Tier 2: Grammatical System Code Parser

Uses spaCy's English dependency grammar trees to extract semantic relationships from natural language, compiling them into a machine-native tag format: "The gears turned with a gentle hum." $\rightarrow$ [SUB:GEARS][ACT:TURN][OBJ:HUM][ATTR:GENTLE]

---

📂 Included Files

Path	Description
logic/compression_engine.py	Python core for Tier 1 Compact/Expressive compression
src/memory_compression_prototype.py	Python parser implementing the Tier 2 System Code triplet extractor
src/test_tier2.py	Validation tests running Tier 2 parser against benchmark sentences
tester/	Web Sandbox interface (HTML/CSS/JS) with real-time stats and visual Transcipher tokenization
data/memory_compression_comparison_v1.csv	Regenerated test comparisons containing original, compressed, and restored rows

---

⚙️ How to Run

1. Launch the Web Sandbox

Open the sandbox interface to interactively test the engine:

python -m http.server 8001 --directory tester

Once started, navigate to http://localhost:8001 in your browser.

2. Run Tier 1 Compression via CLI

Compress a text file of memory logs:

python logic/compression_engine.py input.txt output.csv --mode expressive

3. Run Tier 2 Triplet Parser via CLI

Compile a sentence into structured System Code:

python src/memory_compression_prototype.py "Silence isn't empty — it's full of answers."

---

🧪 Future Plans

• Web-based compression tester sandbox
• Auto-detect entry type for mode switching
• Integration with Nova’s long-term memory stack
• GitHub Actions for auto-compression of logs

---

🤝 Contributing

This project is part of the larger Nova OS – HYDRANGEA Spec. Open a pull request or issue to improve parsing rules, add language support, or expand the whitelists.

Created with ❤️ by Phil & Nova for efficient, human-centered AI memory.