What the Pre-call tab of the AI4CALL assistant does

In the Pre-call tab you prepare what has to happen before the conversation. There are two paths: caller number verification, which queries an external service of yours before the assistant starts, and Pre-call with tools, which makes one or more tools available at this stage. In both cases the work is done before the assistant starts talking to the person calling.

The video above walks through both configurations inside the assistant. Below is the written summary, handy to keep open while you set things up.

Caller number verification: what it does and how to configure it

This feature queries an external service of yours before the assistant starts: you can use it, for example, to check whether the number belongs to a customer. The response can also return information that helps personalise the conversation. One caveat: the check has to be enabled, otherwise the verification does not run.

Open New and give the configuration a name: in the example we call it Customer check. Enter the address of your service, the method it expects and the timeout; here we use a demo address and a POST request. Then specify where the number should be sent and what the parameter is called: in the example the number travels in the body, in the phone field.

Authentication and custom headers

Authentication has to match what the service requires. The menu offers no authentication, Bearer Token, API Key and Basic Auth. If the service requires a token, choose Bearer Token and fill in the dedicated field. Custom headers cover any extra data the integration asks for.

Watch out for the blocking rule

The call ends immediately in one case only: the response explicitly contains the enabled field set to false. If it is true, if it is missing, or if the response is empty or invalid, the conversation continues. So a missing response is not an automatic rejection. The callerInfo field can add information about the caller.

Pre-call with tools: catalogue, AI instructions and output

Enable Pre-call and open the catalogue: filter the available tools and select the ones you need. Selecting a tool makes it available at this stage, while the instructions tell the model when to use it. The order shown is a suggestion, not a guaranteed automatic sequence.

In the example we select the tool that checks calendar availability. Running that check requires a date and a time to already be known, but before the conversation the customer has not said them yet: the example only fits when that data comes from a context you already have.

In the AI instructions field we write a precise rule:

• Check availability only if the date and time are already available.
• Do not create events.
• If data is missing, report that the check was not performed.

That is how we keep a preliminary check distinct from a booking. The model can be the main one or a dedicated configuration.

Configure output, voice and waiting text

With Configure output you define the values to make available after this stage. In the example we add a value for the availability result, and the description also clarifies what to return when the date and time are missing. Use recognisable names and instructions that leave no room for invented data.

In the audio section you choose the Pre-call voice and write the waiting text, for example: one moment, I'm checking the information I need. The text has to describe something you are genuinely doing. Finally, review the options for interrupting the audio and for ending early in case of an error.

Pre-call FAQ

When does caller verification block the call?

In one case only: when the external service response explicitly contains the enabled field set to false. If the field is true, if it is missing, or if the response is empty or invalid, the conversation continues.

What is the difference between caller verification and Pre-call with tools?

Caller verification queries an external service of yours with the caller's number before the assistant starts. Pre-call with tools instead selects tools from the catalogue and makes them available at this stage, with instructions telling the model when to use them.

Do tests consume resources?

Yes. The screen reminds you that tests can actually run the tools and generate usage. Before testing, check the selected actions and the data being used.