Book a callDE Deutsch

Every module.One picture of risk.

Use one module or all of them. They share people, policies and the audit trail, and each one is also an API.

Email security

Reads every message after your mail filter and removes what got through. Each verdict shows its reasons, and each action can be undone.

An invoice from a lookalike domain is flagged with three reasons and removed from 14 inboxes.

  • Microsoft 365 and Google Workspace, no MX change

  • Reported mail lands in one triage queue

Simulations and training

Phishing, SMS, QR and voice simulations in German and English. Whoever clicks gets a two-minute lesson on the spot, and a follow-up lesson after it.

The report of a parcel simulation sent to 180 people: 41 reported it, and the 9 who scanned it saw a short lesson.

  • Email, SMS, QR, voice and chat

  • Sensitive lures need a recorded approval

AI data protection

Haldir Guard masks customer names, prices and drawing numbers in the browser before a prompt leaves the device, and restores them in the answer.

A prompt about a quote is sent with placeholders for the customer, the price and the drawing number, and the answer shows the real values again.

  • ChatGPT, Claude, Copilot, Gemini, Mistral and more

  • Rolled out through Intune or your MDM

AI tool inventory

AI apps, MCP servers and their tools in one list. Set each tool to allowed, approval required or blocked, and agents calling through Haldir get that answer before the call runs.

The GitHub MCP server opens to its tools, delete_repository is switched to Blocked, and the coding agent's next delete_repository call is denied by policy and logged.

  • Per-tool permissions for every MCP server

  • Unknown AI apps found by Haldir Guard

Agent security

Every AI agent gets an owner, scopes and a policy. Tool calls that move money or data wait for approval, and every decision is logged.

An invoice agent's payment to an unknown IBAN is held and denied by its owner.

  • MCP servers and n8n flows

  • In preview

Verification

An urgent payment or access request by phone, video or chat is checked with the real colleague on a number you already know. Their answer arrives in Haldir, and a no alerts security.

A caller posing as the CEO asks for a transfer. The employee checks with the real CEO, who answers in Haldir that it was not her.

  • Teams, phone, email and chat

  • Every check in the audit trail

Data leaks

Each finding shows the work account, the breach it appeared in and what leaked. Ask the person to change the password, then confirm the finding is resolved.

A work account found in a combolist: IT asks for a password reset, then marks the finding resolved. No password is shown or stored.

  • Reset request and resolution are separate steps

  • The account and what leaked, never the password

Device protectionin preview

The device agent is in preview and not released yet. It is designed to stop malicious editor extensions and packages before they install, and to hold packages published too recently.

Preview with sample data: on one laptop a fake AI coding extension is blocked as malware, and an npm package published five days ago is held.

  • In preview, not yet available

  • The console shows sample data until the agent ships

Risk and reporting

One explainable score for people and agents, with reports for the board, your auditor and the works council.

  • NIS2, DORA and AI Act Art. 4

  • Pseudonymous by default