# How do I catch a tool that starts failing more often than it used to?

That's what Tessary's tool_error classifier is built for: it tracks each tool's own failure rate over time rather than looking at individual calls, and fires when the rate shifts away from that tool's normal pattern.

Every tool gets its own baseline, built from its own history, and an alert threshold set against a false-alarm budget rather than one fixed number applied everywhere. Evidence builds up call by call, so ordinary noise, an occasional failure at the tool's usual rate, stays quiet, and only a sustained change crosses the bar. That's the distinction that matters here: an API that starts refusing requests, a dependency that breaks, or a rate limit that begins to bite all show up as a rate change long before every call is failing.

When it fires, the finding names when the shift began and which call sites saw it, so triage can confirm it's real before it reaches anyone as a case.

---

Source: https://tessary.ai/answers/tessary-tool-error/how-do-i-catch-a-tool-failing-more-than-it-used-to
More on Tessary tool error: https://tessary.ai/answers/tessary-tool-error
From Tessary, agent reliability for AI agents in production: https://tessary.ai