Inputs are how you pass information to an Agent when you start a Run. They are the key to making your agents reusable and dynamic, allowing you to run the same agent with different data for a wide range of tasks. By defining a clear input structure, you can build powerful agents that fit your workflows and are easy to trigger programmatically.

​
Defining the Input Schema

Every agent has an Input Schema, which tells it what kind of data to expect for a run. This schema is written using JSON Schema, a popular and standardized way to define the structure of JSON data. Let’s imagine we’re building a “Sales Lead Profiler” agent. We want to give it the name of a person and the company they work at, and optionally some other details. Here’s what the input schema would look like: 
{
  "type": "object",
  "properties": {
    "name": {
      "type": "string",
      "description": "The person's full name"
    },
    "company": {
      "type": "string",
      "description": "The company the person works at"
    },
    "jobTitle": {
      "type": "string",
      "description": "The person's job title (optional)"
    },
    "website": {
      "type": "string",
      "description": "Company website URL (optional)"
    },
    "linkedinUrl": {
      "type": "string",
      "description": "Public LinkedIn profile URL (optional)"
    }
  },
  "required": ["name", "company"]
}
This schema tells us:
  • The input must be a JSON object.
  • name and company are required.
  • jobTitle, website, and linkedinUrl are optional.
When you create a run for this agent in the Ag.dev Console, a form will be automatically generated based on this schema, ensuring that you provide valid inputs every time.

​
Using Inputs in Your Goal Prompt

The real power of inputs comes from using them to dynamically control your agent’s behavior. You can reference the properties from your input schema directly in your agent’s Goal Prompt. We support Handlebars syntax for templating, so you can use {{ propertyName }} for placeholders and {{#if}}...{{/if}} for conditional logic. Here’s how we could write the goal prompt for our Sales Lead Profiler: 
Create a concise sales lead profile for {{ name }} at {{ company }}.
{{#if jobTitle}}Their job title is {{ jobTitle }}.{{/if}}
{{#if website}}Review the company website ({{ website }}) for recent news or initiatives that could be relevant.{{/if}}
{{#if linkedinUrl}}Also consider any public information from their LinkedIn profile: {{ linkedinUrl }}.{{/if}}

Please return a few bullet points covering their background, responsibilities, recent company initiatives, potential buying triggers, and a recommended outreach angle.
Now, when you start a run with an input like this: 
{
  "name": "Dana Lee",
  "company": "Acme Robotics",
  "website": "https://acmerobotics.com"
}
The agent will receive a perfectly tailored goal prompt: 
Create a concise sales lead profile for Dana Lee at Acme Robotics.
Review the company website (https://acmerobotics.com) for recent news or initiatives that could be relevant.

Please return a few bullet points covering their background, responsibilities, recent company initiatives, potential buying triggers, and a recommended outreach angle.
This allows you to create a single, highly-capable agent that you can reuse for any new lead, simply by changing the inputs.
We use Handlebars for templating, which means common blocks like {{#if}}, {{#unless}}, and {{#each}} are all supported.