stream_object

Stream structured output asynchronously with real-time validation and partial object updates.

Overview

stream_object provides real-time streaming of structured output with automatic validation. It combines the benefits of streaming (low latency) with structured output (type safety) by providing both text chunks and partially-parsed objects as they become available.

Basic usage

1. Kick off the stream

   from ai_sdk import stream_object, openai
   from pydantic import BaseModel
   
   class Weather(BaseModel):
       location: str
       temperature_c: float
   
   result = stream_object(
       model=openai("gpt-4.1-mini"),
       schema=Weather,
       prompt="Report the current temperature in Berlin as JSON.",
   )
   
2. Consume deltas + partial objects

   async for delta in result.object_stream:
       print(delta, end="")
   

   Pass on_partial=lambda obj: print("partial", obj) to receive partially-parsed objects while streaming.

Parameters

Same as stream_text plus:

Name	Type	Required	Description
schema	Type[BaseModel]	✓	Pydantic model defining the desired output shape.
on_partial	Callable[[BaseModel], None]	–	Callback executed when a partial object is successfully parsed.

Return value

stream_object returns a StreamObjectResult with:

• object_stream: Async iterator yielding text chunks
• object(): Async method to get the complete parsed object
• text(): Async method to get the complete text
• usage: Token usage statistics
• finish_reason: Why the stream ended
• tool_calls: Tool calls if any were made

Examples

Basic streaming with objects

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel

class Story(BaseModel):
    title: str
    characters: List[str]
    plot: str

async def main():
    model = openai("gpt-4.1-mini")
    result = stream_object(
        model=model,
        schema=Story,
        prompt="Write a short story about a robot learning to paint"
    )

    # Stream the text chunks
    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    # Get the final parsed object
    final_object = await result.object()
    print(f"\n\nFinal object: {final_object}")

asyncio.run(main())

With partial object callbacks

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel

class Recipe(BaseModel):
    title: str
    ingredients: List[str]
    instructions: List[str]

async def main():
    model = openai("gpt-4.1-mini")

    def on_partial(obj):
        print(f"Partial object: {obj}")

    result = stream_object(
        model=model,
        schema=Recipe,
        prompt="Create a recipe for chocolate chip cookies",
        on_partial=on_partial
    )

    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    final_recipe = await result.object()
    print(f"\n\nComplete recipe: {final_recipe}")

asyncio.run(main())

With system instructions

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel

class Product(BaseModel):
    name: str
    price: float
    description: str
    category: str

async def main():
    model = openai("gpt-4.1-mini")
    result = stream_object(
        model=model,
        schema=Product,
        system="You are a helpful product catalog assistant. Always provide accurate product information.",
        prompt="Create a product description for a wireless headphones"
    )

    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    product = await result.object()
    print(f"\n\nProduct: {product}")

asyncio.run(main())

With custom parameters

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel

class Poem(BaseModel):
    title: str
    verses: List[str]
    theme: str

async def main():
    model = openai("gpt-4.1-mini")
    result = stream_object(
        model=model,
        schema=Poem,
        prompt="Write a poem about the ocean",
        temperature=0.8,
        max_tokens=300
    )

    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    poem = await result.object()
    print(f"\n\nPoem: {poem}")

asyncio.run(main())

With complex nested schemas

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel
from typing import List, Optional

class Address(BaseModel):
    street: str
    city: str
    country: str

class Contact(BaseModel):
    email: str
    phone: Optional[str] = None

class Person(BaseModel):
    name: str
    age: int
    addresses: List[Address]
    contact: Contact

async def main():
    model = openai("gpt-4.1-mini")
    result = stream_object(
        model=model,
        schema=Person,
        prompt="Create a person profile with multiple addresses"
    )

    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    person = await result.object()
    print(f"\n\nPerson: {person}")

asyncio.run(main())

Error handling

stream_object handles validation errors gracefully:

import asyncio
from ai_sdk import stream_object, openai
from pydantic import BaseModel, ValidationError

class User(BaseModel):
    name: str
    age: int

async def main():
    model = openai("gpt-4.1-mini")

    try:
        result = stream_object(
            model=model,
            schema=User,
            prompt="Create a user with invalid data"
        )

        async for chunk in result.object_stream:
            print(chunk, end="", flush=True)

        user = await result.object()
        print(f"\n\nUser: {user}")

    except ValidationError as e:
        print(f"Schema validation failed: {e}")

asyncio.run(main())

Tool-calling with streaming objects

See the dedicated Tool page for a complete walkthrough.

import asyncio
from ai_sdk import tool, stream_object, openai
from pydantic import BaseModel

class Calculation(BaseModel):
    result: float
    operation: str
    steps: List[str]

add = tool(
    name="add",
    description="Add two numbers.",
    parameters={
        "type": "object",
        "properties": {"a": {"type": "number"}, "b": {"type": "number"}},
        "required": ["a", "b"],
    },
    execute=lambda a, b: a + b,
)

async def main():
    model = openai("gpt-4.1-mini")
    result = stream_object(
        model=model,
        schema=Calculation,
        prompt="Calculate 15 + 27 and explain the steps",
        tools=[add],
    )

    async for chunk in result.object_stream:
        print(chunk, end="", flush=True)

    calculation = await result.object()
    print(f"\n\nCalculation: {calculation}")

asyncio.run(main())

---

stream_object is provider-agnostic. Swap openai() for anthropic() or any other future implementation – no code changes required.