GitHub - perplexityai/numbat: Visibility into AI agent activity on endpoints, with on-device detection, optional pre-action blocking, and forensic reconstruction.

Open original

Numbat monitors AI agent actions on endpoints using local hooks, OTLP/HTTP logs and on-disk artifacts. It applies CEL-based detection and optional pre-action blocking, outputs versioned NDJSON records for events, findings and enforcement decisions, and offers forensic reconstruction and investigation tools.

  • Numbat monitors AI agent actions on endpoints via local hooks and OTLP/HTTP logs, running CEL-based detection entirely on-device.
  • Blocking is off by default—it only works with specific synchronous pre-action hooks and requires explicitly setting enforce: true plus installing numbat in enforce mode.
  • It can forensically reconstruct agent activity from on-disk session artifacts even if numbat wasn't installed at the time, using read-only commands like agents, scan, and timeline.
  • Ships as a single dependency-free binary (no cgo) for macOS, Linux, and Windows, with commands for hook management, rule testing, and building portable investigation case bundles.

numbat brings unified endpoint visibility for AI agents across desktop apps, CLIs, IDEs and gateways. It hooks into supported agents via local plugins and OTLP/HTTP log exporters, then normalizes live and at-rest activity into a single event model. A built-in Common Expression Language (CEL) engine runs detection locally; you can write results to stdout, a file or push them over HTTP. Blocking is off by default and works only with specific synchronous pre-action hooks—enforcement requires marking rules with enforce: true and installing numbat in enforce mode. Under the hood, numbat ships with multi-step sequence rules, custom YAML rules and JSON Schema–defined NDJSON output for events, findings, enforcement decisions, indicators and scan summaries. It can reconstruct forensics from on-disk session artifacts even if numbat wasn’t installed beforehand. You’ll get read-only commands—agents, scan, timeline—to map what’s present and timeline multi-step actions. Hooks live-capture only start in monitor-only mode; to turn on blocking you copy a shipped rule into your policy directory, set enforce: true, bump the version and run numbat hook install with --enforce. The single-binary works on macOS, Linux and Windows (amd64, arm64) and builds without cgo. Download releases with SHA-256 checksums or install via go install …/numbat@latest. For custom builds: CGO_ENABLED=0 go build -trimpath -o numbat ./cmd/numbat. Once installed, commands include hook install/status/uninstall, scan, collect, ship for record delivery, plus rules check/list/test for policy work, and case build/verify for portable investigation bundles. Everything pivots around live-capture hooks, local CEL evaluation and NDJSON records you can ship or bundle.

Ask about this article