Securing Agents

Different types of AI agents require different security approaches. This page maps each agent category to the right Highflame integration path and links to the detailed documentation for each.

---

Choosing your path

Agent type	Deployment model	Integration
Custom agents — new builds	In-code, per-step control	Highflame SDK
Custom agents — existing traffic	Centralized proxy, no code changes	Agent Gateway
Code agents (Cursor, Claude Code)	Local daemon, IDE hooks	Code Agents
Browser agents (ChatGPT, Gemini, Copilot.ai)	Managed browser extension	Browser Security
Third-party / uncontrolled agents	Network-layer via Tailscale Aperture	Tailscale

These are complementary. Most enterprise deployments use more than one: the Gateway for custom agent traffic, Code Agents for developer workstations, Browser Security for employee AI usage, and Tailscale Aperture for any third-party systems that fall outside those categories.