Agent Discovery

Discover and manage every AI agent running in your organization, whether it writes code on a developer's laptop, runs inside a business platform, or has not been registered anywhere yet, so you can govern it and apply security policy.

What is Agent Discovery

Agent Discovery is the part of Agent Identity Security that answers one question: which AI agents actually exist in my environment, and who owns them? It exists because agents show up in three very different ways, and each way needs its own connection method.

Discovery path When to use it Learn more
Coding Agents The agent runs on a developer's own workstation (Claude Code, Cursor, GitHub Copilot CLI, and similar tools). Installed manually or pushed at scale through your MDM. Coding Agents
Non-Coding Agents The agent runs on a remote enterprise platform (Salesforce Agentforce, n8n, Microsoft Copilot Studio, and similar tools). Connected with an API credential; no software to install. Non-Coding Agents
Shadow Agent Discovery Nobody has registered the agent yet. Agent Identity Security finds it indirectly through your existing EDR or MDM tooling. Shadow Agent Discovery

Choosing the right discovery path

Use the following guidance to decide where to start:
  • If you are rolling out AI coding assistants to your engineering team, start with Coding Agents.
  • If your business teams already use an AI agent platform such as Agentforce, n8n, or Copilot Studio, start with Non-Coding Agents.
  • If you suspect agents are already running that nobody has registered, start with Shadow Agent Discovery; it requires no new integration if you already run an EDR or MDM tool.

All three paths feed the same Registry, so an agent discovered through Shadow Agent Discovery today can be formally onboarded through a Coding Agent or Non-Coding Agent integration tomorrow, without losing its history.

After an agent is discovered

Once an agent is discovered through any path, Agent Identity Security can also:
  • Bind it to a real identity, see Identity Providers.
  • Route its sensitive actions for human approval, see Human-in-the-Loop.
  • Give it, or restrict it from, calling AppViewX capabilities as tools, see MCP.