FAQ
Short answers to the questions security reviewers and evaluators ask most about Holistic AI EndLayer: privacy, enforcement, deployment and hosting.
Short answers about Holistic AI EndLayer, the Guardian Agent Platform. Each one links to the page with the detail.
Evaluating EndLayer
What does EndLayer do?
It finds the AI tools, agents, MCP servers and AI websites in use on enrolled devices and in your connected cloud accounts, grades them against your policy, and stops what your policy says should be stopped. One agent, the Layer 1 Agent, does the work on each device. See the overview.
Who is it for?
Any organisation where people use AI at work and the company needs to see what is in use, what is shared with it, and what it costs. It suits organisations of any size.
Which AI tools does it recognise?
22 AI tools by name, including Claude Code, Cursor, Codex CLI, GitHub Copilot, Windsurf, Gemini CLI, ChatGPT Desktop and Ollama, plus any you add to your own catalogue. It also reads AI website use per browser profile and every declared MCP server. See agents.
Privacy and security
Does EndLayer read what people write to AI tools?
Conversation content is off by default, on the device and in the organisation. Turning it on needs a stated lawful basis, only administrators can read it, and every read is logged with a reason before anything is shown. The content of tool calls is never collected. See privacy and retention.
Does the agent run with administrator rights?
It runs as the person whose device it is and never elevates. Installing it needs administrator rights once. See the trust model.
What happens to secrets it finds?
Secret values are discarded on the device. EndLayer records that a key exists and where it is exposed, never the key itself. See what is collected.
Where is our data hosted?
In Holistic AI's hosted service by default; the Trust page lists the regions, certifications and subprocessors. Self-hosting, including air-gapped networks, is available by arrangement. See where it runs.
Enforcement
Does it block anything?
Yes. Guardrails warn on, redact or block sensitive data in Claude Code, Cursor, Codex CLI and on AI websites. Blocking a tool requests its removal, which an administrator approves. Kill switches halt a tool across the targeted devices, each with a typed reason. Approval before an agent acts is planned. See what it stops.
Does it catch prompt injection?
Yes, in two places. Files uploaded to AI websites are checked on the device before the site receives them, for hidden instructions written for the model. The Guardian plugin in Claude Code, Cursor, Codex CLI and Antigravity warns on or blocks prompt injection and jailbreak attempts. See what it stops.
Which browsers does the extension support?
Chrome, Edge, Brave, Vivaldi and Chromium. Firefox support is planned.
Deployment
Which operating systems are supported?
macOS, Windows and Linux. Phones and tablets are listed through Microsoft Intune. See deployment.
Does it work with Intune or Jamf?
Yes. Both can push the agent from the console to the groups you pick, with results read back per device. SCCM and any other device management tool can use the generated install scripts. See integrations.
Can a person turn the agent off?
On managed Windows devices, and on Macs with the login-items profile in place, the agent keeps running. Any device that stops reporting shows in the console, so a gap is visible either way.
Does it send alerts to our SIEM?
Alerts go to Slack, Microsoft Teams, ServiceNow and email today. SIEM destinations, including Splunk, Microsoft Sentinel and Datadog, are planned. See the roadmap.
Does it map to frameworks?
Yes. Nine built-in frameworks, including SOC 2, the EU AI Act, ISO/IEC 42001, NIST AI RMF and GDPR, each with coverage by control point and an export for auditors. The mappings support a review; they do not certify. See the console.
Updated 2026-11-20