The Start Node: Inputs and State Variables

Typed inputs, explicit state

A workflow begins at a start node that defines the inputs for a run (typically the user's message, sometimes extra variables such as a customer ID or language) and the state variables the workflow will keep. Define them explicitly with sensible types and defaults. Avoid stuffing everything into one free-text message: a separate customer_tier or locale variable is easier to route on, test and log than something an agent has to infer from text. Product details such as node names and menus change; check the current OpenAI documentation.

Start, agent, tools, end

Most workflows combine a start node with inputs, agent nodes, tools and an end node with a defined output.

An input and state design

Name and type everything that routing depends on.

inputs:
  input_as_text: string          # the user message
  customer_id:   string          # from your app, not from the user
  locale:        string = "en"
state:
  category:      "billing" | "technical" | "other" | null
  attempts:      number = 0
  approved:      boolean | null

Pass identity from your app

Supply user or customer IDs from your authenticated backend, never from the chat text, so a user cannot claim to be someone else.

Quick check: Why define routing-relevant values as explicit variables?

• Variables make the model larger
• They are easier to route on, test and log than values inferred from text
• Agents cannot read text
• It removes the need for an end node