Skip to content
Customer Pro

AI agents

Custom responses

Structured output schemas that force an agent to answer in a predictable shape.

On this page

Normally an agent answers in prose. A custom response makes it answer in a shape you defined — a typed structure with named fields — so another system can consume the result instead of a human reading it.

Audio
Fixed answers for questions you always get

When you'd use it #

Classification

Sort every inbound message into a case type and an urgency, and act on the field rather than the sentence.

Extraction

Pull a name, a date and a quantity out of a rambling message into three clean fields.

Routing

Have the agent decide which team a conversation belongs to, and return that decision as a value you can branch on.

Defining a schema #

  1. Name the response

    Something that says what it is for — triage_result, booking_details.
  2. Declare the fields and their types

    Each field needs a name, a type and a description. The description is instruction to the model, so write it as one.
  3. Constrain wherever you can

    A field that must be one of five values will be one of five values. A free-text field will be anything.
  4. Attach it to an agent

    Set it as the agent's custom response in the builder.

Where to go next #