# numbat

numbat gives visibility into what AI coding agents do on a developer's machine, watching supported desktop, CLI, IDE and gateway agents through local hooks, logs and session files, and can optionally block a risky action before it runs.

numbat observes supported desktop, CLI, IDE and gateway AI agents through local hooks and plugins, OTLP/HTTP logs, and on-disk session artifacts, normalizing both live and past activity into one event model evaluated by a shared CEL rule engine. Detection runs locally, and records can be written to stdout, a local file, or delivered over HTTP.

Built-in and custom YAML rules, including multi-step sequence rules, flag suspicious activity, and pre-action hooks can optionally block an action, though enforcement is off by default and limited to rules explicitly marked to enforce. It can also reconstruct a forensic timeline from existing on-disk session artifacts without prior instrumentation, redacts secrets from its output, and ships as a single binary for macOS, Linux and Windows.

Website: https://github.com/perplexityai/numbat