Agents Gone Wild? Use Tool Call Limits in LangChainJS to Keep Them in Check!

In this tutorial, Christian Bromann will show you how to prevent runaway tool usage using the Tool Call Limit Middleware in LangChainJS. You’ll learn how to set clear, declarative limits on tool usage — per run, per thread, or per tool — without wiring up your own state machine or guardrails. We’ll build a real example: 📱 An SMS-sending agent with credits. The user has 2 SMS credits, each message costs 1 credit, and the agent must respect that. What You’ll Learn - How the tool call cycle works in the ReAct agent loop - Where middleware fits inside the agent execution pipeline - How to enforce per-run and per-thread limits Why This Matters? Tool Call Limit Middleware is perfect for: - Protecting against expensive API overuse - Enforcing business logic like user credits or rate limits - Preventing runaway tool loops in agents - Keeping your apps stable and predictable All code shown in the video is available here: 👉 https://github.com/christian-bromann/langchat Read more about ToolCallLimit Middleware in our docs: 👉 https://docs.langchain.com/oss/javascript/langchain/middleware/built-in#tool-call-limit