OpenRouter Agent SDK 0.11.0 adds doom-loop safeguards

openrouter.ai11 min read

AI agents can become trapped in expensive infinite loops when a repeatedly failing tool call provides no instruction to stop, causing costs to escalate rapidly as conversation history grows. The OpenRouter Agent SDK introduces built-in safeguards including doom-loop detection, argument normalization via loopKeys, and per-run step and cost caps. Additionally, API key credit limits provide a server-side defense to halt runaway spending even if agent code fails. Open article →