Skip to main content

How It Works

The AI Assistant is a conversational layer over the platform. This page explains the principles behind it — not the internal implementation, but what you can rely on as a user.

Natural language, not commands​

There is no command syntax to learn. You describe your intent in your own words, and the assistant maps it to a real platform action. If a request is ambiguous, the assistant asks a clarifying question with a selection card rather than guessing.

Real, permission-scoped actions​

Everything the assistant does is a real platform operation — the same one available in the dashboard, performed through the same APIs.

  • Permission-scoped: The assistant inherits your account's roles and permissions. It can only do what you are allowed to do. If you can't do something in the dashboard, the assistant can't do it for you either.
  • Auditable: Because actions run through the standard platform APIs, they are subject to the same logging and audit trail as any other operation.
info

If the assistant declines an action, it is usually because your account lacks the required permission — not because the feature is missing. Ask an administrator to adjust your role. See User Management.

Confirmation & safety​

For actions that change state or cannot be undone, the assistant shows a confirmation or a summary card before proceeding. Destructive or irreversible operations are never performed silently.

This means you stay in control: you can always review what's about to happen and cancel.

Multilingual​

The assistant understands and replies in your language. You can switch languages simply by writing in a different language — no setting to toggle.

Context-aware conversations​

Multi-turn flows keep their context. When you start something that takes several steps — for example, "create a calendar" followed by "confirm" — the follow-up messages are understood in the context of what you were doing, so short replies like "yes", "the second one", or "confirm" land in the right place.

Manual vs scheduled actions​

The assistant can act in two ways:

  • Manual (immediate): Actions you ask for happen right away — opening a stream, starting a recording, launching a job.
  • Scheduled: When you set up a calendar or schedule, execution is handed to Cossa (Cordatus Smart Scheduling Agent), which starts and stops tasks on time.
warning

Cossa checks schedules every minute. If you schedule something for the current minute, it may start on the next cycle rather than instantly. For example, a task set at 14:30:35 may begin at 14:31:00.

For the full scheduling model, see Record Scheduling.

What it does not do​

  • It does not bypass your permissions.
  • It does not perform destructive actions without confirmation.
  • It does not replace the dashboard — it's an additional way to control the same platform.