OpenClaw Rate Limit Fix: Stop 429 Errors for Good

Why OpenClaw Hits Rate Limits

OpenClaw's gateway detects 429 responses from API providers and puts them into cooldown. A known bug can misclassify non-rate-limit errors as rate limits, locking out all your models even when the APIs are healthy. Running a single model with no fallbacks means one 429 kills your entire agent.

The Immediate Fix

Restart the gateway to clear all cooldown timers: openclaw gateway restart. If that doesn't work, do a full reset: openclaw gateway stop && sleep 3 && openclaw gateway install --force && openclaw gateway start.

Configure Fallback Models

This is the real fix. Add models across multiple providers so OpenClaw auto-rotates when one hits limits. Run openclaw models fallbacks add openai/gpt-4o and openclaw models fallbacks add google/gemini-2.5-flash-preview. Use at least 3 different providers (Anthropic, OpenAI, Google) — each has independent rate limits.

Reduce Context Token Burn

Every message sends your entire conversation history to the model. At default settings, that can be 200k tokens per request. Cut it to 50k: openclaw config set agents.defaults.contextTokens 50000. This alone can reduce rate limit hits by 75%.

Fix Telegram Polling Stalls

If you're on macOS, your Mac sleeping drops the Telegram long-poll connection. The gateway then burns API credits in reconnection bursts. Fix: run caffeinate -d & to prevent sleep, or move the gateway to an always-on VPS or Raspberry Pi.

Run the Doctor

Run openclaw doctor --fix to auto-repair common configuration issues that cause rate limit problems. This resolves roughly 70% of gateway issues automatically according to community analysis.