OpenClaw Performance Optimization Guide

Why Optimization Matters

Running an AI agent 24/7 requires careful optimization. Without it, you’ll burn through tokens and experience lag.

5 Key Optimization Techniques

1. Thinking Mode

{
  "thinkingDefault": "minimal"
}

• Reduces response time from 2.2s to 1.1s
• Use “minimal” for real-time interactions

2. WebSocket Throttle

# Reduce throttle for faster responses
WebSocket throttle: 150ms → 20ms
# 7x faster voice response!

3. Prompt Caching (v2026.2.0+)

• 90% cost reduction
• Cache common prompts

4. Model Routing

• Use cheap models for heartbeat
• Reserve expensive models for complex tasks
• Example: Haiku → Sonnet → Opus

5. Context Window Management

• Set contextTokens: 80000 limit
• Regular compaction
• Reserve 24000 tokens floor

Cost Comparison

Setup	Monthly Cost
Default	$87/month
Optimized	$27/month
Savings	70%

Conclusion

Optimization is essential for production AI agents. Start with these 5 techniques!

{
  "thinkingDefault": "minimal"
}

# 减少节流以获得更快响应
WebSocket 节流: 150ms → 20ms
# 语音响应快 7 倍！