How to Optimize OpenClaw Performance and Reduce Costs?

Optimizing OpenClaw involves two dimensions: response speed and API cost reduction.

Speed optimization: - Choose a fast model: GPT-4o-mini and Claude Haiku have the lowest latency. DeepSeek can be slow under load. - Reduce context size: Keep soul.md concise. Fewer words = faster processing. - Limit skill count: Each active MCP server adds startup time. Install only skills you use. - Use SSD storage: Memory retrieval is faster on SSD than HDD. - Upgrade RAM: If the server is swapping, everything slows down. Monitor with htop.

Cost optimization: - Set token budgets: Limit max tokens per request and per day. openclaw config set limits.maxTokensPerRequest 2000. - Use cheaper models for simple tasks: Route simple queries to GPT-4o-mini or Haiku, save Sonnet/Opus for complex ones. - Optimize soul.md: Every token in soul.md is sent with every request. A 1000-word soul.md at 50 requests/day costs real money. - Memory cleanup: Remove outdated memories that pad the context unnecessarily. - Disable unused skills: Skills that respond with 'no relevant tool found' still consume prompt tokens.

Monitoring: - Check the Dashboard for daily token usage breakdown - Set up alerts for spending thresholds on your provider's dashboard - Use openclaw stats to see cost trends over time

Typical costs: Light personal use $3-5/month, moderate use $10-15/month, heavy business use $20-50/month.

Tip: Run openclaw stats weekly to spot cost trends. A sudden spike usually means a skill is making excessive calls.

# View usage stats
openclaw stats

# Set daily token budget
openclaw config set limits.dailyTokenBudget 100000

# Set max tokens per request
openclaw config set limits.maxTokensPerRequest 2000