Just saw a feature request for model circuit breakers in the OpenClaw repo (issue #55536). This is something I have been running externally for months and it changed everything.
The Problem
When your LLM provider starts failing — rate limits, 503 errors, degraded quality — OpenClaw keeps retrying the same broken endpoint. You get cascading errors and your entire session dies.
What a Circuit Breaker Does
Same pattern web services use for database failover:
- Model fails 3 times consecutively → circuit OPENS (model disabled)
- Requests auto-route to healthy alternative
- After 5-minute cooldown → circuit HALF-OPEN (test request)
- If test succeeds → circuit CLOSES (model re-enabled)
- If test fails → stay open, try again in 5 minutes
Why This Matters
- No more cascading failures when one provider has issues
- Work never stops — automatic failover to alternatives
- You stop burning rate limit retries on a provider that is clearly down
My Setup
I use TeamoRouter which handles circuit breaking across Claude, GPT-4o, DeepSeek, Gemini, and MiniMax. When Claude rate-limits me, traffic shifts to DeepSeek automatically. When Claude recovers, traffic shifts back.
Free tier available with unlimited MiniMax M2.7 calls (teamo-free mode).
Discord for failover configs and circuit breaker setup help.
Top comments (0)