Sage

Sage is an open source security layer for AI agents from Gen Digital, the company behind Norton. It sits between an agent and its tools, inspecting each tool call before it runs and stopping dangerous commands, malicious URLs and credential leaks. Gen describes it as agent detection and response. When an agent issues a bash command, a URL fetch or a file write, Sage extracts the artifacts involved, such as URLs, shell commands, file paths and package names. It checks them in parallel against local YAML heuristics, cloud URL reputation checks and package supply chain checks, then returns a verdict of allow, ask or deny. Results are cached and written to an audit log. The categories it covers are dangerous commands such as reverse shells and destructive deletes, malicious URLs, credential leaks including API key patterns and SSH key access, supply chain attacks such as typosquatted npm and PyPI packages, obfuscation and evasion tricks, and persistence or tampering such as cron jobs, shell profile edits and attempts to disable Sage itself. It installs as a plugin for OpenClaw, Claude Code, Cursor and VS Code, and the page says it needs no further configuration once installed. The code is published on GitHub. The site lists no prices.