# Overview

> What Holistic AI EndLayer governs on your devices and in your cloud accounts, what it measures and stops, and where to read more.

Holistic AI EndLayer, the Guardian Agent Platform, answers a question most organisations find hard to answer about themselves: **which AI tools are running on our devices, how are they set up, what do they cost, and does any of it match our policy?**

AI tooling arrives through people, not procurement. Somebody installs a coding agent, points it at a repository and gives it a key for an internal service. None of that shows up in an application inventory, a licence report or a purchase order. EndLayer makes it visible, grades it against your policy, and stops what your policy says should be stopped.

EndLayer is for any organisation where people use AI at work, of any size.

## Measure, grade, act, prove

  - [Measure](https://www.endlayer.ai/docs/how-it-works): The Layer 1 Agent reports what AI is installed on each device and how it is set up.
  - [Grade](https://www.endlayer.ai/docs/how-it-works): Facts are graded against your policy, and anything without a policy stays ungraded.
  - [Act](https://www.endlayer.ai/docs/enforcement): Guardrails warn, redact or block in real time, and kill switches halt a tool.
  - [Prove](https://www.endlayer.ai/docs/console): Every verdict keeps its evidence, and every decision is in the audit trail.

## One agent on every enrolled device

The Layer 1 Agent runs on macOS, Windows and Linux. It has three jobs: it **finds** the AI on a device, **reads** how it is configured and used, and **stops** what your policy names. It scans every 15 minutes, runs as the person whose device it is, and never runs the tools it finds.

Phones and tablets are listed through Microsoft Intune, so your inventory covers them too.

## What it measures

Each scan reports 26 signals in seven groups. The full list is on the [signals page](https://www.endlayer.ai/signals).

| Group | The question it answers |
|---|---|
| **Tools and agents** | What is installed, and can it act without asking? |
| **Apps and web** | Which AI services are used, including through a browser? |
| **MCP and access** | What has an agent been given access to, and what does it hold? |
| **Accounts and spend** | Whose account is it, and what is it costing? |
| **Projects and data** | Which repositories and data files are agents working in? |
| **The device** | What is the machine, and is it managed and up to date? |
| **Security posture** | Are the controls a security team already asks about in place? |

EndLayer recognises 22 AI tools by name, including Claude Code, Claude Desktop, Codex CLI, ChatGPT Desktop, Cursor, Windsurf, GitHub Copilot, Gemini CLI, Antigravity, Cline, Continue, Aider, Goose, Ollama and LM Studio. You can add your own to the catalogue. It also reads AI website use per browser profile and every MCP server declared in a configuration file.

## What it stops

- **Sensitive data in prompts:** 31 built-in detectors plus your own patterns, in Claude Code, Cursor, Codex CLI and on AI websites through the browser extension.
- **Prompt injection in documents:** files uploaded to AI websites are checked on the device before the site receives them.
- **Agent behaviour:** the Guardian plugin and five computer use and impersonation rules.
- **Unwanted software:** blocking a tool requests its removal, which an administrator approves.
- **Kill switches:** halt a tool, a desktop app's website, an MCP server, a device group or the whole fleet, each with a typed reason.

See [what it stops](https://www.endlayer.ai/docs/enforcement) for each control.

## Beyond the device

Connections add what a device cannot see: managed devices that never enrolled, the people behind them, and AI that runs in your cloud accounts. EndLayer connects to Microsoft Intune, Entra ID, Agent 365, Microsoft 365 Copilot, Copilot Studio, Defender and Purview, Jamf, SCCM, Google Cloud and Vertex AI, Amazon Bedrock AgentCore and Microsoft Foundry, and sends alerts to Slack, Microsoft Teams and ServiceNow. See [integrations](https://www.endlayer.ai/docs/connections).

> **Available today, and planned:** Everything on these pages is available today unless it is marked as planned. Planned work includes approval before an agent acts, SIEM destinations, SDKs for your own applications and Firefox support. The [roadmap](https://www.endlayer.ai/roadmap) lists each item.

## Where to go next

  - [How it works](https://www.endlayer.ai/docs/how-it-works): The agent, the browser extension, the backend and the console.
  - [Trust model](https://www.endlayer.ai/docs/trust): What the agent can do, and the promises built into it.
  - [Deployment](https://www.endlayer.ai/docs/deployment): One device in minutes, or a fleet with no one at any of them.
  - [See it for yourself](https://www.endlayer.ai/docs/see-it): Look at a populated console before enrolling a device.

Source: https://www.endlayer.ai/docs/overview · Updated 2026-11-20
