v1.2.5 — Adaptive Forgetting + Multi-Model Decay

What's New

Multi-Model Decay System

• 4 decay models now available:
  • exponential — Original model (default)
  • power_law — Better for long-term stable memories
  • step_function — Periodic checkpoint decay
  • forgetting_curve — Ebbinghaus-inspired, practice-dependent
• Configurable via MEMORY_DECAY_MODEL env var or per-memory via memory_set_decay_model

Type-Aware Default Models

• fact → power_law (long-term stability)
• experience → exponential (rapid forgetting)
• preference → step_function (stable preferences)
• skill → forgetting_curve (practice-dependent)

Adaptive Forgetting

• MEMORY_ADAPTIVE_DECAY=true enables dynamic decay rate adjustment
• Tracks retrieval success rate per memory
• High-retrieval memories automatically slow their decay
• Low-retrieval memories accelerate toward archival

Forgetting Analytics

• memory_forgetting_report — View decay rates, retrieval stats, and adjusted rates
• memory_set_decay_model — Change decay model per memory

Schema Migration

• v10 → v11: Added forgetting_analytics table
• Added decay_model and decay_params columns to memories table

New Environment Variables

• MEMORY_DECAY_MODEL — Global default decay model (default: exponential)
• MEMORY_ADAPTIVE_DECAY — Enable adaptive forgetting (default: false)

New Tools

• memory_forgetting_report — Forgetting analytics dashboard
• memory_set_decay_model — Per-memory decay model override