Your data, your rules.

Each company has its own schema in the database and its own database role. Shared tables are protected by row-level security. Uploaded files are stored in the company's own folder. The company is identified by its address, such as yourcompany.core3.digital.

On managed Kubernetes in an OVHcloud data centre in Warsaw. The YugabyteDB database runs on three nodes.

Permissions live in the core of the system, not in individual screens. Roles and grants decide which records a person works with. Policies only narrow access. Field rules hide, lock or mask a specific value, such as a salary. The same rules apply to screens, reports, exports, automations and AI.

Settings that apply to the whole platform, such as maintenance mode, file storage and technical limits like attachment size, are changed only by the operator of core3. A company can see them but cannot change them, and both the application and the database itself enforce this.

core3 records who changed what and when, including attempts the permissions stopped, and report runs and exports. Only people an admin gives the right can see the history, and only for their own company. Records cannot be edited or deleted: the company's database role may only read and add them.

ClamAV scans every file before it is saved. A file with a virus is not saved, and neither is a file the scan could not check. Only people whose permissions allow it can download a file.

Browsers talk to the server only over HTTPS, with HSTS on. Sign-in cookies are encrypted and available only to the server. Sensitive settings are stored encrypted in the database: mail credentials, Microsoft and Google sign-in and AI keys.

The admin picks the provider

AI comes with every plan. It answers through the provider and model your admin picks: the one we offer, or your own key, for example for OpenAI or Anthropic. With your own key, your terms with the provider apply and nothing is taken from the core3 budget.

It reads with your permissions

The assistant sees only what the person asking may see. A write, such as a new meeting, is shown first and saved only once confirmed.

Fields kept from the AI

A field rule hides or masks, say, salary or IBAN for the assistant only, even when the person can see it. The rules don't come preset; your admin sets them.

What goes to the provider

The question, the data the assistant reads for the answer, and attachments you send it. The content of the open page only if you consent.

Budget

The company has a monthly budget from its plan. Once it runs out, prepaid credit that never expires is used, or pay-as-you-go up to a cap, if you switch it on. An admin divides the budget, with a limit for a role or for a specific person. Once a limit is spent, the assistant starts no further steps.

Memory

Off until an admin switches it on. Everyone sees what the assistant remembered and can delete it. It never remembers passwords, account or card numbers, e‑mails, phone numbers or national ID numbers.

Notifications and helpdesk replies go out from your Microsoft 365 or Google Workspace mailbox, or through your SMTP server with DKIM signing.

Reports download to Excel, CSV or PDF. The export runs on the server, always with the permissions of the person who started it. There is no public API yet.

Send security questions and vulnerability reports to core@igt-systems.com.