Hey Product Hunt 👋 I built Agent Watchdog after watching my own LangGraph pipelines quietly burn through tokens — two agents debating in circles, or a graph stuck waiting on a condition that never resolves. Standard tracing tools (LangSmith, AgentOps) are great for post-mortems, but I wanted something that intervenes while it's happening, not after. It watches message similarity and tool-call patterns across a sliding window, and halts the run the moment it detects an argument loop, a repeated tool call, or a stuck iteration — before you're staring at a $40 API bill for a conversation going nowhere. Just shipped a new version with tool-call loop detection and state-regression checks. Would love feedback, especially from anyone running multi-agent systems in production — what failure modes have you hit that this doesn't catch yet?

Agent-watchdog

Runtime loop detection for LangGraph agents

Runtime loop detection for LangGraph agents