Perplexity AI open-sources Numbat agent security tool
Perplexity AI has released Numbat, an open-source security tool built in Go that monitors, detects, and optionally blocks risky behavior from AI coding agents on local endpoints. Operating entirely on-device, Numbat evaluates agent activity using a Common Expression Language (CEL) rule engine, captures live hooks and logs, and supports retrospective session reconstruction from on-disk artifacts without compromising developer privacy.
As autonomous AI coding agents gain broader shell execution privileges, client-side runtime security is transitioning from an optional safeguard to essential enterprise infrastructure.
- –Local deterministic detection avoids latency penalties while maintaining strict developer data privacy.
- –Configurable enforcement modes allow organizations to audit agent behavior safely before turning on active blocking rules.
- –Retrospective artifact scanning gives security teams forensic capabilities to audit past agent execution timelines.
- –Cross-harness compatibility ensures uniform security policy enforcement across different AI coding assistants and CLI tools.
DISCOVERED
1h ago
2026-08-03
PUBLISHED
1h ago
2026-08-03
RELEVANCE
AUTHOR
0x4D31

