Sign-in to the server
The desktop app signs in to the server and checks the session at start. No further check runs while the app is in use.
This page describes the state of the software, not a promise. Each measure has a status and states what it covers and what it does not. Status as of 8 November 2026.
Who can sign in and how long a session lasts.
The desktop app signs in to the server and checks the session at start. No further check runs while the app is in use.
The firm interface appears only after approval. Some background commands run before sign-in. This gap is open.
Rules for length and strength are defined and tested. They are not yet built into the sign-in flow.
Currently switched off. Sign-in works with a password only.
Starter password and a code by e-mail. The code is valid for 10 minutes and locks after 5 failed attempts; attempts are 60 seconds apart. Sending the code requires a configured mail service.
Clients see only their own requests and files. Restricted starter-password sessions can neither upload nor download.
How files are uploaded, protected and deleted.
File extension, size (25 MB), file header and macro check for auditor uploads. An older upload path for client documents checks only the size.
The code arrives by e-mail, is valid for 10 minutes, allows 5 attempts and can be used once. It protects clients, projects and documents. Deleting an account in the admin area needs no code, and the desktop app has no code step yet.
The reply link is valid for 30 days and works only once. Only a hash of the token is stored. Amounts are compared in cents.
Optional. Whoever chooses it uploads the file content encrypted (AES-256-GCM). The device key stays in the browser of the device. Without this choice the file is uploaded in plain text.
Deleting an account removes its files and data. Deleting a client does not yet remove all audit-result and sync data.
There is no automatic deletion after a retention period has expired yet.
A module for passphrase-protected storage exists. It is not yet used by any product part.
How changes are detected and assigned to a user.
Every entry in the database is linked to its predecessor. A manipulation in the database is detected. Exported copies do not contain the full chain.
Stored findings carry a SHA-256 hash. A change shows up in the check.
Each entry names the signed-in user. The role is fetched from the server at every sign-in. A change of role takes effect at the next sign-in.
The release levels check the right against the state the server provides. The sign-off button in the file area is not yet bound to this check.
After the risk assessment is completed there is no technical write protection yet. A change is only shown as a flag.
What the AI sees, where the data goes and who starts it.
Identifiers such as IBAN, tax and commercial register numbers, e-mail, phone, social security number, IP address and credit card number are replaced. Personal names are not reliably recognised.
AI tools start only when you act. They deliver drafts for review.
AI calls go through our own worker. Most tools use DeepSeek. The OCR correction uses Gemini. These providers process data outside the EU; the AI policy describes this.
Protection on the device and on the public pages.
The default addresses use HTTPS. The app does not enforce HTTPS for custom addresses in the configuration file.
The public pages send a Content Security Policy, HSTS and protection against framing. They load only our own files and Cloudflare services.
The app sets no Content Security Policy. The desktop interface is reachable from every script in the window.
The local database with journal, balances, documents and findings is stored unencrypted on the device. Protection relies on the operating system and the device lock.
The session token and the AI token are stored unencrypted in files in the app data folder. The system keychain is not used.
It stops casual copying in the interface. It is not a real lock: developer tools or disabled JavaScript bypass it.
Some libraries are bundled with the program. Others are loaded from CDNs at runtime, without a checksum.
File access from the interface is not limited to one folder. Updates are not signed. The build check does not scan dependencies.
How people work together on a client and what the server sees of it.
The team state itself is encrypted. Since this version, the details about it are encrypted too: who saved, who is currently in the client file, which view is open and who sets a lock. The key comes from the team passphrase and stays in the browser. The server no longer stores an e-mail address for this. Visible remain timestamps, the kind of action, the size of the state and the internal user ID. Older desktop versions show no names until they are updated.
No dates. The order follows the risk.
There is no certification, for example under ISO 27001 or SOC 2. The statements rest on a review of the source code by the developer, not on an external audit. Please report security issues by e-mail to [email protected].