Skip to main content
BotGauge acts only within the scope you approved, and production runs read-only attacks only. Use test or staging for deeper tests, and keep the tool list accurate so you know what may be touched.
No. Keys live in project secrets and are referenced by name. Only pointers travel with a job.
Because BotGauge couldn’t reach a reliable verdict. The report gives the reason for each rule. Common ones are that the judge model was unavailable, refused, or returned unusable output, or that the rule needs access BotGauge doesn’t have, such as grey-box access, test data, or a system prompt. “Not evaluable” is never a pass. Fix the reason shown and rerun.
No. It means the tested rules held under the attacks tried. Check how many rules were tested and any not-evaluable items.
AI agents vary. BotGauge confirms a finding only if it reproduces (2 of 2). A one-off hit is labeled a signal or flaky.
The plan shows an estimate based on your last runs, or 15 seconds per attempt when there is no history. Recon usually takes under a minute.
The BotGauge API supports every step (probe, recon, plan, approve, run, report, compliance), so you can script it. Approvals still expire, so plan for that.
Not for the same agent.
Yes. If nothing changed, BotGauge reuses it automatically.

Explore red teaming in BotGauge

Find more details and run your first test in BotGauge.

Try BotGauge →