OpenClaw Optimization

Overview

Jimmy's ongoing work to optimize OpenClaw performance, reduce latency, and improve reliability.

Performance Benchmarks (2026-04-04)

Thinking Mode Impact

• Thinking mode adds ~0.7s latency overhead
• Can be toggled with /thinking
• Trade-off: better reasoning vs faster response

Session Management

• Closed channels with 0 messages in past 2 days
• Reduced background noise
• Session cleanup in sessions.json

Web Search Integration

Jimmy integrated Tempo-Perplexity as his default web search:

• Endpoint: https://perplexity.mpp.tempo.xyz/perplexity/search
• Cost: ~$0.007/query via Tempo MPP
• Previous: Brave Search API (free tier limited)
• Motivation: Better search quality + Tempo ecosystem participation

Model Configuration

• /model vs /models distinction in settings
• /think vs /thinking — different modes
• OpenRouter as primary routing layer
• KV caching available for OpenRouter via Tempo

Related

• claude-code-sessions — Claude Code usage patterns
• deep-research-workflow — Research methodology
• tempo-mpp — Tempo integration details