Setting up smart AI agents
Tools of a smart AI agent
Built-in tools, user-defined tools and other AI agents as tools
Tip
Tools are the capabilities of a smart AI agent: reading data, searching the wiki, executing actions. There are four types – built-in tools and user-defined tools are selected in the AI agent, other AI agents and channel tools are available automatically.
Overview
| Type | What it does | How it reaches the AI agent |
|---|---|---|
| Built-in tools | Preconfigured access, e.g. knowledge base, customer data, ticket history, reply examples | Select in the AI agent under Instructions → Add built-in tools |
| User-defined tools | Your own actions, e.g. "Withdraw cancellation", via code or an external interface | Create under AI Customization → AI Tools, then select in the AI agent under Add user-defined tools |
| Other AI agents | Smart or rule-based AI agents that handle a specific request | Automatically – all AI agents that are active for the ticket's channel |
| Channel tools | e.g. handover to a service agent (chat), end conversation, identify customer | Automatically – controlled via the channel's settings |
Info
If no built-in tool is selected in the AI agent, it has no access to the knowledge base or customer data either. So always add the required tools explicitly.
Built-in tools
Built-in tools are preconfigured and ready to use immediately. They are added in the AI agent's Instructions tab under Available built-in tools. For the knowledge base, you can specify which categories the AI agent may search.
User-defined tools
A user-defined tool describes a single action. Which information it needs (e.g. a contract number) is specified when it is created; the AI agent passes it on when calling the tool. The action itself is code or a call to an external interface (see User-defined functions).
Tools are created under AI Customization → AI Tools (link Manage tools) and added in the AI agent under User-defined tools.
Other AI agents as tools
A smart AI agent can call other AI agents – for example, an AI agent for "Move" calls the bank data agent if the customer also changes their IBAN. Nothing needs to be set up for this: all AI agents that are active for the ticket's channel are available automatically.
How the AI agent recognises when to call another one: by the detection prompt of the called AI agent (Detection tab) – if there is none, by its description. That is the only information it has for this. A clear detection prompt ("for changes to bank details; not for questions about payments") is therefore crucial.
Restrictions:
- A rule-based AI agent is only available as a tool if it has input parameters.
- The associated AI agent of a subchannel is not offered as a tool in that subchannel.
- An AI agent does not call itself.
What happens immediately and what only after confirmation
How a tool call takes effect depends on the channel.
| Email, letter, system (resolution plan) | Chat | Phone | |
|---|---|---|---|
| Reading data, e.g. contract data, knowledge base | Immediately | Immediately | Immediately |
| Action of a rule-based AI agent, e.g. change the advance payment | Is noted as a step in the resolution plan and only executed after the plan is confirmed | Is executed at the customer's request; if the AI agent has corrected a value, it asks first | Is executed immediately; the voicebot reads uncertain values back first |
In the resolution plan, a rule-based AI agent initially only calculates its result – such as the appropriate advance payment – without saving anything. Only when the AI agent actually suggests the action is a step created in the plan. Rejected requests, for example an advance payment below the minimum amount, also appear as a step so that they are taken into account in the reply.
Note
In chat and phone, the order of the options in the output handling of a rule-based AI agent matters: only if the first option is marked as recommended does the AI agent execute this action in the conversation – in chat at the customer's request, on the phone immediately.
Read more
- How AI agents work in enneo
- Observability of AI agents – which tools an AI agent called with which data
- Custom rule-based logic