ktg-memory-compress

name

ktg-memory-compress

description

Context Extension Protocol (CEP) for compressing conversation context into portable carry-packets. Use when conversation approaches context limits (80%+ tokens), user says "save context", "compress this", "create carry packet", "memory compress", or when switching models/sessions. Implements Progressive Density Layering (PDL) with 4 layers achieving 6:1 compression while preserving semantic fidelity. Output is JSON carry-packet for cross-session/cross-model context restoration.

KTG Memory Compress (CEP v4.1-INTRA)

Context Extension Protocol for compressing conversation context into portable carry-packets.

Core Concept

CEP transforms sprawling conversation context into dense, structured JSON restorable in new sessions or different models. Compression: 6:1. Forensic recall: 9.52/10 average.

Triggers

• Auto: Context at 80%+ (~160K tokens)
• User: "save context", "compress", "carry packet", "memory compress"
• Session end: Before long conversation terminates
• Model switch: Preparing context for different LLM

Progressive Density Layering (PDL)

Layer	Allocation	Contents
L1: Knowledge	40%	Definitions, key decisions, facts + confidence
L2: Relational	25%	Dependencies, conflicts, resolutions
L3: Contextual	20%	Reasoning archetypes, domain principles
L4: Metacognitive	15%	Technique effectiveness, session fingerprint

Compression Workflow

1. Assess Context (RKQDE)