v0.5.0 - Progressive TOON Compression

🚀 Features

Progressive TOON Compression (Recommended)

Intelligent checkpointing that compresses history only when beneficial:

• Trigger 1: Context size ≥ 80% of model window (prevents truncation)
• Trigger 2: Time gap ≥ 5 minutes (prevents expensive cache miss recovery)
• Savings: 94% vs no optimization

Config:

{
  "historyCompression": {
    "enabled": true,
    "progressive": {
      "enabled": true,
      "contextThreshold": 0.8,
      "timeThreshold": 300
    }
  }
}

Token Tracking

• Estimates tokens before/after TOON compression
• Displays savings in /router usage

Cache Hit Display

• Shows 📦 with cached token count
• Format: ↑5,230 ↓2,140 📦125,000 $0.0234

Static TOON Mode

• Freeze TOON at specific turn (freezeAfter: 10)
• Savings: 92% vs no optimization

📊 Cost Comparison (100-turn session)

Strategy	Total Cost	Savings
No optimization	$22.50	0%
Cache only	$2.95	87%
Dynamic TOON	$11.00	51%
Static TOON	$1.90	92%
Progressive TOON ⭐	$1.35	94%

🐛 Fixes

• Corrected freeze condition logic
• Restored safeSplit calculation
• Fixed type safety issues

⚙️ Backward Compatibility

All features opt-in. Default behavior unchanged.

Full Changelog: v0.4.0...v0.5.0