NHI Inventory

Every agent and identity, each with an owner

NHI Inventory is one register of every AI agent and the service accounts, keys and tokens it relies on: four pillars built as one system, each entry with a named owner.

Agent inventory
Discovery on devices
Discovery in your directory
Discovery sources

At a glance

What's in NHI Inventory

Why it mattersEvery other control depends on knowing which agent is actingThe register Agent Authority works from
Agent inventoryOne register of every agent, each with a named ownerAgent register
Discovery on devicesThe agents nobody registered, on the machines people useShadow agents
Discovery in your directoryRisky app and agent identities, reviewed on a scheduleDirectory review · What each entry shows
Discovery sourcesThe identity, endpoint and network tools you already runDiscovery sources · Providers

Part of Agent Authority

The register the rest of Agent Authority works from

A rule that only knows the user can't tell one agent from another using the same credentials. Knowing which agent is acting, and who owns it, comes first.

MCP, LLM and Endpoint Authority decide on the agent, not only the person it acts for
Agent Threats & Findings ties each finding to an agent and its owner

See how it fits together on the Agent Authority overview.

AgentClaude CodeActive

Owner: Platform team
Purpose: Code review on pull requests
Allowed tools: GitHub, Atlassian

MCP AuthorityWhich agent is calling a tool
LLM AuthorityWhich agent is calling a model
Endpoint AuthorityWhich agent runs on a device
Agent Threats & FindingsWhose agent a finding is on

Pillar 1 · Agent inventory

One register of every agent, each with a named owner

Every other control depends on knowing which agent is acting. NHI Inventory keeps one register of every agent in use, along with the service accounts, API keys and tokens those agents rely on, surfaced and mapped.

Each agent is a named, versioned, registered non-human principal
An accountable owner, a stated purpose and a list of allowed tools
Map each one to an owning team

The Agent Instances tab of Agents & NHIs in the Agent Authority console: totals for registered agents, agents active in the last 24 hours and tool executions in the last 24 hours, above a table of Codex, Claude and Claude LLM instances, each with its runtime and version, tags such as Coding, ChatGPT and Claude, an Active status, and when it was last active and created. The user under each agent is blurred. — From the Agent Authority console

Pillar 2 · Discovery on devices

Find the agents nobody registered, on the machines people use

Sensors installed on each laptop, desktop or CI runner are listed with their owner, when they were last seen and whether they are active or revoked. Agents found there that were never registered show up in the register too.

Early access: sensors report the AI agents people run, such as Claude and Claude Code, and flag processes that are not in the catalog
The number of devices and users for each, so you get a real count of unmanaged AI use

Local MCP servers, where each agent sends data and on-device rules are covered in Endpoint Authority.

The Shadow Agents tab of Agents & NHIs: 16 agents seen, 3 in the catalog and 13 unrecognised, with filters for all, in catalog and not in catalog. The table lists Claude and Claude Code, recognised by their own process, and svchost.exe, Code and start-hyprland marked Not in catalog because no signature matched, each with the AI domains it called such as api.anthropic.com, its device and user counts, and when it was last seen. — From the Agent Authority console

Pillar 3 · Discovery in your directory

Surface the risky app and agent identities in your directory

App registrations and agent identities keep multiplying, and they often sign in with long-lived secrets and no MFA. Agent Authority reviews them in your configured directory, such as Microsoft Entra, on a schedule.

Early access: scheduled reviews of the app registrations and agent identities in your configured directory
Findings are listed by severity, with the rule, the identity and its kind, service principal or application
Each finding explains why it was flagged and what to do next, and can be acknowledged or dismissed

The findings themselves live in Agent Threats & Findings.

A finding in the Agent Authority console: Tenant-wide admin consent, severity High, status Open, on a Microsoft Copilot Studio agent. Why flagged: an application an admin consented to on behalf of everyone in the tenant. What to do: review the tenant-wide consent grant and revoke it if it is broader than needed. Below, the identity (kind Agent identity, its external ID, source Entra) and when the finding was first and last seen. — From the Agent Authority console

The Findings tab of Detections in the Agent Authority console: what the gateway and your connected directories flagged as worth a look, filtered to Status: Open and sorted by severity. Each row shows when it was last seen, its severity, all Critical here, the rule, such as Disabled by Microsoft or Secret never expires, the identity, which is blurred, its kind, Service principal or Application, the status Open, and buttons to acknowledge or dismiss it. — From the Agent Authority console

Pillar 3 · Discovery in your directory

Every directory agent, with its IDs, owners and open findings

The agents your connected directory knows about, such as Microsoft Copilot Studio agents in Microsoft Entra, are listed with when they were created and how many findings are open. Open one to see the rest.

Object type, object ID, app ID and blueprint
Whether it's enabled in the directory, and when it was created
Its owners and sponsors, so an agent with no owner stands out
Open findings with their severity

The NHI Agents tab of Agents & NHIs: 19 directory agents across 4 products, all 19 with open findings, the highest High. The table lists Microsoft Copilot Studio agents with when each was created and its open findings, such as 1 open, Medium. The first row is expanded to show object type Agent identity, its object ID, app ID and blueprint, In directory Enabled, the created date, Owners None, and its sponsors, which are blurred. — From the Agent Authority console

Pillar 4 · Discovery sources

Discover NHIs from the tools you already run

Connect the identity, endpoint and network providers you already use. The agents and non-human identities they know about land in one register.

Agent providers bring agent telemetry in
NHI sources list the non-human identities in your directories
Each source shows whether it's authenticated and when it was last checked

The NHI Sources tab of Discovery Sources in the Agent Authority console, with an Add NHI source button and one provider, Entra (Microsoft Entra), whose status is Authenticated and which was last checked 1 minute ago. — From the Agent Authority console

Discovers from

Microsoft Entra
Okta
CrowdStrike
Zscaler
F5
And more

Identity, endpoint and network providers you already run, each added as a source under Discovery Sources.

Related capabilities

NHI Inventory FAQ

What is NHI Inventory?

NHI Inventory is one register of every AI agent and the service accounts, API keys and tokens those agents rely on, each with a named owner. Agents are found on devices and in your connected directories, and anything not in your catalog is flagged.

Which non-human identities does NHI Inventory keep track of?

Every agent in use, as a named, versioned, registered non-human principal, along with the service accounts, API keys and tokens those agents rely on. From your directory it also lists app registrations and agent identities, such as Microsoft Copilot Studio agents in Microsoft Entra.

How does NHI Inventory find agents nobody registered?

In early access, sensors installed on laptops, desktops and CI runners report the AI agents people run, such as Claude and Claude Code, and flag processes that are not in the catalog. Each comes with the number of devices and users, so you get a real count of unmanaged AI use.

What does NHI Inventory check in my directory?

In early access, Agent Authority reviews the app registrations and agent identities in your configured directory, such as Microsoft Entra, on a schedule. Each finding explains why it was flagged and what to do next, and you can sort by severity to fix the biggest exposures first.

Which tools can NHI Inventory discover from?

Connect the identity, endpoint and network providers you already use, such as Microsoft Entra, Okta, CrowdStrike, Zscaler and F5, and more. Each source shows whether it's authenticated and when it was last checked, and the agents and non-human identities it knows about land in one register.

How does NHI Inventory fit into Agent Authority?

It is the register the rest of Agent Authority works from. A rule that only knows the user can't tell one agent from another using the same credentials, so knowing which agent is acting, and who owns it, comes first. The findings themselves live in Agent Threats & Findings.

Know every agent and who owns it

14-day free trial. Your own Agent Authority tenant.Add your first source under Discovery Sources, and its agents and non-human identities start filling the register. Read how the agent registry works and how non-human identities are surfaced.

Start free trial