Create a new agent when you want part of your business to speak with a different tone, different instructions, or a different scope of knowledge than the rest of your agency — for example, an agent specialized in rentals, with its own system prompt and access only to the documents tagged for that part of the business (section 6). **Requirements:** Admin role. 1. Tap **Create agent**, at the top right of the screen. 2. A window opens with the same fields as the editor. Type the **agent's name** (required). 3. Adjust the **LLM model** and the **tone** if you don't want to leave them at their defaults (**Automatic** and **Neutral**). 4. Type the **system prompt** (required) — the agent's base instructions: who it is, its goal, and how it should behave. It has a character counter, up to a maximum of **8000**. 5. Type the **fallback message** (required) — what the agent replies when it doesn't know how to answer something. 6. Add **escalation rules** if you need them (optional; field-by-field detail in section 4). 7. Adjust its **resources** if you don't want the agent to start with no access to anything (optional; section 6). 8. Tap **Create agent** to confirm, or **Cancel** to discard the window. **Result:** the agent shows up immediately in the list on the left, **Active** by default and ready to be assigned to a channel (**Manual 24 · Channels**). > **Note — Fixed transparency note under the system prompt** > > Both in this window and in the editor (section 3), right below the system prompt field, the platform > always shows this notice: "Transparency (EU AI Act, Art. 50): the platform already tells your customers > they're talking to an AI — a notice on the first contact of every channel and on AI↔human switches — and > the agent will confirm it's an AI if asked. These guarantees can't be turned off; make sure your own > instructions don't ask it to do the opposite." It isn't an editable field — it's fixed information so you > know what your own prompt shouldn't contradict.