Repolode
Submit
perplexityai

perplexityai/numbat

Endpoint visibility into AI agent activity, with local detection, optional pre-action blocking, and forensic reconstruction from session artifacts.

View on GitHub
Stars
1,162
Forks
125
Open issues
14
Last commit
4 days ago

What is it?

numbat observes supported desktop, CLI, IDE, and gateway agents through local hooks and plugins, OTLP/HTTP log exporters, and on-disk session artifacts. Live and at-rest activity is normalized into a single event model and evaluated by the same CEL rule engine, with detection running locally. Records can be written to stdout or a local file and optionally delivered over HTTP.

Blocking is off by default and applies only to supported synchronous pre-action hooks and to rules explicitly marked for enforcement. All shipped rules are monitor-only, so operators must copy a rule, set its enforce flag, and install that policy themselves. Forensic reconstruction works from supported session artifacts without prior instrumentation, and read-only scans redact secrets from output.

The project ships as a single binary built without cgo for macOS, Linux, and Windows. It includes inventory, per-session timelines, and portable case bundles with SHA-256 manifests.

Key features

  • 01Live monitoring through hooks, plugins, and OTLP/HTTP log exporters.
  • 02Local detection with built-in CEL rules, multi-step sequence rules, and custom YAML rules.
  • 03Opt-in blocking at supported pre-action hooks, disabled by default.
  • 04Forensic reconstruction from supported on-disk session artifacts.
  • 05Versioned NDJSON records with JSON Schemas for the wire format.
  • 06Read-only artifact scanning with secret redaction.
  • 07Portable case bundles with SHA-256 manifests.

Use cases

  • Monitoring what coding agents read, write, and execute on developer workstations.
  • Investigating an agent session after the fact from its on-disk artifacts.
  • Blocking a narrow set of high-risk actions, such as writes to SSH authorized_keys, after a deliberate policy is set up.
  • Sharing an incident's evidence as a verifiable case bundle.

Quick start

  1. 1

    Install numbat with Go 1.26.9 or newer

    go install github.com/perplexityai/numbat/cmd/numbat@latest
  2. 2

    List discovered agents and scan their artifacts (read-only)

    numbat agents
  3. 3

    Install live monitoring hooks for Codex (monitor-only by default)

    numbat hook install --agent codex --emit all

Things to know

  • Live monitoring coverage varies by agent and surface, so check the coverage matrix before relying on it.
  • Blocking is off by default and all shipped rules are monitor-only; enforcement requires writing and maintaining your own rules.
  • Hooks need host trust review after installation and after upgrades, which adds operational steps.

In collections

Open-source AI agent frameworks & apps

Topics

#agent-security#ai-agents#endpoint-security#threat-detection#audit-logging#forensics#cel#go#ai-agent-security

FAQ

What is numbat?
Endpoint visibility into AI agent activity, with local detection, optional pre-action blocking, and forensic reconstruction from session artifacts. numbat observes supported desktop, CLI, IDE, and gateway agents through local hooks and plugins, OTLP/HTTP log exporters, and on-disk session artifacts. Live and at-rest activity is normalized into a single event model and evaluated by the same CEL rule engine, with detection running locally. Records can be written to stdout or a local file and optionally delivered over HTTP.
Is numbat free to use? What license does it have?
numbat is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.
Is numbat still maintained?
Actively maintained. The last commit to numbat was on Nov 4, 2026.
What can I use numbat for?
Monitoring what coding agents read, write, and execute on developer workstations. Investigating an agent session after the fact from its on-disk artifacts. Blocking a narrow set of high-risk actions, such as writes to SSH authorized_keys, after a deliberate policy is set up. Sharing an incident's evidence as a verifiable case bundle.
How do I install numbat?
Install numbat with Go 1.26.9 or newer: go install github.com/perplexityai/numbat/cmd/numbat@latest List discovered agents and scan their artifacts (read-only): numbat agents Install live monitoring hooks for Codex (monitor-only by default): numbat hook install --agent codex --emit all
Where can I get numbat?
The source code and installation instructions are on GitHub: https://github.com/perplexityai/numbat

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects