Perplexity open sources Numbat, a security layer for AI coding agents

A MacBook with lines of code on its screen on a busy desk

AI coding agents are running more autonomously on developer machines every month. Perplexity AI just shipped a tool designed to keep an eye on what those agents are actually doing.

What Numbat Is

Perplexity AI open sourced Numbat on July 29 to 30, 2026. It’s a lightweight agent-detection and response layer that monitors AI coding agent activity across client endpoints. The project is distributed under the Apache 2.0 license and was contributed to the Open Secure AI Alliance.

What It Does

Numbat operates at two points in the agent lifecycle. Before an agent acts, it provides local, pre-action runtime guardrails. After something goes wrong, it supports post-incident forensics. Both run locally on the developer’s machine, covering macOS and Linux.

The configuration uses YAML with CEL (Common Expression Language) for defining rules, and NVIDIA hardware is referenced in the supported stack.

Why It Matters for Developers

If you’re running AI coding agents in your local environment, you currently have limited visibility into what those agents touch, modify, or transmit before an action completes. Numbat sits between the agent and the endpoint to surface that activity in real time and log it for review afterward. For solo developers and small teams, that’s a meaningful gap this tool directly addresses.

The Apache 2.0 license means you can integrate it into commercial workflows without restriction. Worth watching if agent-based coding is part of your stack.

Stay on top of AI & Automation with BizStack Newsletter
BizStack  —  Entrepreneur’s Business Stack
Logo