With custom agents, you create AI agents in the Agent Hub for your individual processes – in addition to the built-in AI colleagues such as the Sales order agent or the Returns agent. You describe in your own words what your agent should be able to do, and an assistant creates the agent from this. Your agent then runs automatically in your email workflow: Incoming inquiries on its topics are forwarded to it and are displayed in the Tasks tab as usual.
Just like the built-in agents, your custom agent only prepares actions. They are only carried out when you approve them in the task.
You can find your custom agents on the Home page in the Custom agents area.
Steps:
-
Open the Home page in the Agent Hub.
-
In the Custom agents area, click on Create agent.
The assistant is opened with the question "What should your new agent do?".
-
In the "Message the assistant…" input field, describe what the agent should do, or select one of the suggestions.
The suggestions are based on your recent tasks, for example "An agent that checks delivery status by order number".
The agent editor is opened. On the left, you see the chat with the assistant, on the right the draft of your agent. The assistant creates the name, capabilities, tools, and behavior description and displays what it changed under Updated.
-
If necessary, answer the assistant’s follow-up questions, for example about exceptions or special cases, and review the draft (see Editing a custom agent).
-
Click on Save & test agent to try out the agent on a real task (see Testing a custom agent), or directly on Save & deploy.
Your agent is displayed on the Home page in the Custom agents area and is active.
Tip
Describe one clearly defined use case per agent, for example "An agent that answers customer questions about the grind size and origin of our coffee". The more precisely you describe which inquiries the agent should handle and how it should respond, the better it works.
To edit an existing agent, hover over the agent’s card on the Home page and click on Edit.
On the left, you work with the assistant: In the "Message the assistant…" input field, describe what you want to change about the agent, for example "Always add a reference to our online shop to your replies". The assistant then adjusts the configuration. On the right, you can see the current configuration:
Field name |
Description |
|---|---|
|
Name |
The name of the agent. Use the pencil icon next to the name to change it. |
|
This agent will be used for: |
The agent’s capabilities as short keywords. They control which incoming inquiries are forwarded to this agent. |
|
Model |
The AI language model the agent works with. A short description of the model is displayed below the selection. |
|
Allowed tools |
The tools the agent may use to access data in Xentral. Tools with a pencil icon are write tools: They change real data and count as actions. Use the info icon on a tool to get a description. The Tools and Actions bars display how many tools and actions you have already assigned. |
|
How your agent behaves |
The behavior description of the agent, for example its task, the desired style and tone, and its way of working. Below the field, you can see how many characters you have already used. |
In the header, you also find the following buttons:
-
Expert mode: Here you define tools and capabilities manually (see Defining tools and capabilities in expert mode).
-
Delete: Deletes the agent after a confirmation prompt. Alternatively, you can delete an agent using the trash icon on its card on the Home page.
-
Save & test agent: Saves the agent and starts a test run (see Testing a custom agent).
-
Save & close: Saves your changes and closes the agent. For a new agent, the button is called Save & deploy.
In Expert mode, you define yourself which tools the agent may use and what it is used for.
Steps:
-
In the header of the agent, click on Expert mode.
-
In the Allowed tools area, check the tools the agent may use. Use Search tools… to find a specific tool.
You can find write tools under Actions, for example Cancel order or Create credit note.
-
Next to This agent will be used for:, click on Add capability to add a further capability. Use the X icon on a capability to remove it.
-
Click on Save & close.
Note
You can assign up to 10 read tools and, in addition, up to 5 actions per agent. The behavior description can contain up to 4,000 characters. You can create up to 50 custom agents per instance.
Before your agent handles real inquiries, you can try it out on one of your recent tasks.
Steps:
-
In the opened agent, click on Save & test agent.
The agent is saved. The list of your recent tasks is displayed in the chat.
-
Pick a task to test with. Use the search field to find a specific task.
The agent processes the task as a test. After a short time, the chat displays how it would have handled the task under Suggestion for this task: its proposal, the summary of the customer request, the decision recommendation, and the proposed actions.
-
Click on Done or use Test another task to test a further task.
Important
No actions are carried out in Xentral during the test run. The agent only displays which steps it would propose.
If the result does not fit yet, describe to the assistant what the agent should do differently and test again.
Note
With Save & test agent, the agent is saved and is active right away. If it should not handle real inquiries yet, deactivate it on its card (see Activating or deactivating a custom agent).