OpenClaw Model Fallback Strategy: Keep Agents Online Under Rate Limits

The Core Principle

Fallback should optimize for availability first, then cost and style.

Practical Chain Design

Use cross-provider sequencing:

1. Primary high-quality model
2. Different provider (similar quality)
3. Different provider (faster/cheaper)
4. Long-tail backups

This prevents provider-level incidents from taking down the whole chain.

Operational Rules

• Never place two same-provider models back-to-back when possible
• Probe all models periodically with a tiny health task
• Log model-level failures separately from prompt failures
• Keep one “boring but stable” fallback at the end

Prompt Compatibility Tips

• Keep system prompts concise and provider-neutral
• Avoid provider-specific XML formats unless necessary
• Validate tool-call behavior across all fallback models

Bottom Line

A fallback chain is not just cost optimization. It is your uptime architecture.