When your provider returns a rate limit, OpenClaw puts it in cooldown and retries. Set up fallback models so your bot stays responsive. If rate limits happen often, check your billing console.
Context length errors mean your conversation grew past the model's window. Reduce maxHistoryMessages in your config. The default is ; try - when you see these.
You'll hit two known bugs that misreport errors. Issue #5240 reports any provider failure as "rate_limit". Issue #10368 misreports rate limits as "Context overflow" during failover. When the error doesn't match what you're seeing, check both on GitHub.