Is tool_error turned on by default?

Yes. tool_error runs from the moment a project connects its traces, one of three classifiers Tessary calls deterministic and turns on by default, alongside duration_drift and cost_drift. They’re on at the start because none of them needs a rubric or a threshold you have to pick: a tool call either returned an error or it didn’t. The rest of the built-in set, frustration and secret_leak among them, ships off until you enable it for a project.

You can turn it off per project if a noisy tool makes it unhelpful there. On is the starting position, not a requirement.

Being on by default is not the same as being tuned to your agent. Each tool is judged against its own history, so one that has always failed sometimes sets its own bar, and the first shift tool_error reports may be one your team already knew about.

keep reading

More on this.

Does tool_error count a tool's output as failed just because it mentions the word error?No. A call only counts as failed on structural evidence, an error status, a recorded exception, or a result that itself declares an error, never text.answer →Does tool_error flag a single failed tool call?No. It's built on a tool's failure rate over time; one failed call inside a stable baseline doesn't move the rate enough to cross the threshold.answer →How do I catch a tool that starts failing more often than it used to?Watch each tool's failure rate against its own baseline instead of any single call; that's what Tessary's tool_error classifier does.answer →Everything on tessary tool errorThe definition, and every question answered under tessary tool error.concept →

Send us the traces you already emit.

book a demo