Problem

OpenClaw natively supports and encourages subagent spawning for task parallelism, but session accumulation severely impacts host performance, forcing users to sacrifice data retention for operational stability.

Real-World Metrics (from our deployment)

Before (447 sessions accumulated over ~3 days)

After (aggressive manual pruning to 23 sessions)

Session Breakdown (typical)

• 2-4 long-lived main/channel sessions (webchat, Slack)
• ~400+ ephemeral sessions (subagents, spawn-dispatch, cron jobs)

With dispatcher running every 10 minutes spawning 2 sessions each cycle, accumulation is ~12 sessions/hour × 24h = ~288 sessions/day.

Root Cause

1. sessions.list serializes all sessions on every call — O(n) with expensive per-session work
2. Control UI/Dashboard polls sessions.list frequently (~every 7 seconds), even when not viewing Sessions tab
3. archiveAfterMinutes (default 60) doesn't keep pace with spawn rate
4. No pagination, caching, or incremental sync

Current Workaround

We had to:

1. Manually archive old .jsonl transcripts
2. Run openclaw sessions cleanup --enforce --fix-missing
3. Set archiveAfterMinutes: 120 (still aggressive)
4. Accept losing troubleshooting history to maintain stability

This is a poor tradeoff — we lose the ability to debug issues from hours ago.

Suggested Improvements

1. Pagination for sessions.list — don't serialize 400+ sessions per request
2. Reduce Control UI poll frequency — or skip sessions.list when not on Sessions view
3. Incremental sync — send deltas instead of full list
4. Tiered retention config — separate policies for:
   • Main/channel sessions (keep days/weeks)
   • Subagent sessions (keep hours)
5. Hard session count cap with LRU eviction — auto-archive oldest ephemeral sessions when limit reached
6. Background proactive cleanup — don't wait for user to hit performance cliff
7. Lazy loading — defer full session metadata until requested

Environment

• OpenClaw v2026.3.28
• macOS 15.7.2 (arm64), Apple Silicon
• Workload: Automated dispatcher spawning subagents for issue processing

Impact

This is a significant reliability issue for anyone running automated subagent workflows. The system encourages subagent spawning but doesn't scale session management to match, creating a hidden performance cliff that's hard to diagnose and forces users to choose between history retention and system stability.