Assistant Quick Start# PRO

The Assistant API in Scripting provides three distinct capabilities, each designed for a different type of use case: structured data, streaming output, and interactive conversations.

Before choosing an API, first decide what kind of result you need.

#Assistant API Overview

#requestStructuredData

Purpose Requests strictly structured JSON output that conforms to a provided schema.

Best suited for

• Parsing receipts, invoices, and bills
• Extracting fields from natural language
• Generating configuration or rule objects
• Any output that must be consumed by program logic

Key characteristics

• Stable and predictable output
• No streaming or incremental updates
• Ideal for background or headless scenarios

In one sentence

If you need data, use requestStructuredData.

#requestStreaming

Purpose Requests streaming output, allowing you to receive content incrementally as the model generates it.

Best suited for

• Typing-effect UI
• Long-form content generation
• Low-latency user feedback

Key characteristics

• Emits text, reasoning, and usage chunks
• Can be rendered progressively
• Output is not guaranteed to be structured

In one sentence

If you need real-time output, use requestStreaming.

#Conversation APIs

Related methods

• startConversation
• present
• dismiss
• stopConversation

Purpose Creates and presents a system-hosted Assistant chat experience.

Best suited for

• ChatGPT-style interactions
• Multi-turn conversations
• Scenarios where the system manages UI, streaming, and provider switching

Key characteristics

• Built-in chat UI
• Streaming handled automatically
• Only one active conversation at a time

In one sentence

If you need a full chat experience, use the Conversation APIs.

#How to Choose the Right API

#Common Scenarios

• Parse a receipt → requestStructuredData
• Show AI writing text live → requestStreaming
• Open a chat interface for users → Conversation APIs
• No UI, just results → requestStructuredData or requestStreaming
• Let the system manage the chat UI → Conversation APIs

#Minimal Examples

#Structured Data

const result = await Assistant.requestStructuredData(...)

#Streaming Output

const stream = await Assistant.requestStreaming(...)
for await (const chunk of stream) {
  // handle chunk
}

#Conversation

await Assistant.startConversation({
  message: "Hello",
  autoStart: true
})
await Assistant.present()

#Usage Tips

• Do not mix Conversation APIs with requestStreaming in the same flow
• Prefer requestStructuredData whenever output must be consumed as data
• Use streaming or conversations for presentation-focused scenarios

#Next Steps

For deeper details, refer to:

• requestStructuredData – detailed schema-driven data extraction
• requestStreaming – streaming behavior and chunk handling
• Conversation APIs – lifecycle and interaction patterns