Inputs and Outputs

The inputs and outputs section is used to describe the inputs and outputs of the agent. The inputs and outputs section is typically written in YAML, and is included at the top of the agent definition. Let’s look at this section in more detail.

inputs:
  firstName:
    type: string
    default: User
    description: The first name of the customer
  lastName:
    type: string
    description: The last name of the customer
  question:
    type: string
    description: The question to answer
outputs:
  thing:
    type: string
    description: The thing to output

The inputs and outputs section is broken down into two sections:

• inputs: The inputs to the agent. This is typically a set of inputs that will be used as parameters that participate in the template rendering. Each input is broken down into three sections:

  • type: The type of the input. This can be either a string, number, array, object, or boolean.
  • default: The default value of the input.
  • sample: A sample value of the input. This is used to provide examples to the user of the agent. They can also be used to provide examples to the tooling that will be used to load and execute the agent. If this section is not included, the runtime will use the default value for the input.
  • required: A boolean value that indicates whether the input is required. If this section is not included, the runtime will assume that the input is required.
  • strict: A boolean value that indicates whether the input can emit structural text used when parsing the output. If this section is not included, the runtime will assume that the input cannot emit structural text by defaulting strict to true. If strict is set to false, the runtime will assume that the input can emit structural text.
  • description: A short description of the input. This is typically a short string, but can be any string that is compatible with the agent.
  • [optional] json_schema: A JSON schema that describes the input if the type is object.

• outputs: The outputs of the agent. This represents the expected output type of the agent (the output, however, is governed by the provider - sometimes this section can be ignored). Each output is broken down into two sections:

  • type: The type of the output. This can be either a string, number, array, object, or boolean.
  • default: The default value of the output.
  • description: A short description of the output. This is typically a short string, but can be any string that is compatible with the agent.
  • [optional] json_schema: A JSON schema that describes the input if the type is object.

  The json_schema section is used to specify the JSON schema that describes the input or output. This is typically a JSON schema that is compatible with the input or output type. If this section is not included, the runtime will assume that the input or output is a simple type (e.g. string, number, array, object, or boolean).

A shorthand for the outputs section could be something like:

outputs: ${file:schema.json}

This would expand the json schema into the outputs section.

A note on the inputs, outpus, and tool sections with respect to configuration. The current approach uses a “keyed” naming approach when delcaring inputs, outpus, as well as tool parameters (more on tools below). This approach may not be the desired approach by some providers. This preference will also be honored by allowing an array of properties as well:

inputs:
  - name: firstName
    type: string
    default: User
    description: The first name of the customer
  - name: lastName
    type: string
    description: The last name of the customer
  - name: question
    type: string
    description: The question to answer