📄 compaction-configuration-update.md 1,529 bytes Apr 21, 2026 📋 Raw

OpenClaw Compaction Configuration Update

Date: 2026-04-21
Status: ✅ Applied
Scope: All three agents (main/Wadsworth, Daedalus, Socrates)

Summary

Successfully configured OpenClaw to increase compaction threshold from ~20K to ~100K tokens per your revised directive. This maximizes context utilization for cloud-hosted agents while acknowledging the Sovereign Constraint (local-first architecture).

Configuration Changes (CORRECTED)

1. Global Defaults (agents.defaults.compaction)

{
  "compaction": {
    "mode": "safeguard",
    "reserveTokensFloor": 28000,
    "reserveTokens": 28000,
    "keepRecentTokens": 15000,
    "recentTurnsPreserve": 6,
    "maxHistoryShare": 0.8,
    "memoryFlush": {
      "enabled": true,
      "softThresholdTokens": 5000
    },
    "postIndexSync": "async",
    "qualityGuard": {
      "enabled": false
    }
  }
}

Key Settings:

Setting Value Effect
reserveTokensFloor 28000 Compaction triggers at ~100K tokens (128K - 28K reserve)
reserveTokens 28000 Ensures 28K tokens reserved for prompt generation
keepRecentTokens 15000 Preserves 15K of most recent context during compaction
recentTurnsPreserve 6 Keeps last 6 turns verbatim
maxHistoryShare 0.8 Allows 80% of context for history (vs 50% default)
mode safeguard Stricter guardrails to preserve recent context

Result: Compaction now triggers at ~100K tokens instead of ~20K.

← Back