Easily Stream LLM Responses with Django-Bolt and PydanticAI

I like how easy it is to create an async streaming endpoint with django-bolt and PydanticAI from scratch. With only a few commands you can set it up.

Start a Django project

Use uv to quickly create a new Django project:

# Create a fresh Django project
uvx --from django django-admin startproject my_django_agent
cd my_django_agent

# Install django-bolt and pydantic-ai
uv init --no-package
uv add django django-bolt pydantic-ai

# add django-bolt to INSTALLED_APPS
vim my_django_agent/settings.py

Create an API endpoint

Create a streaming API endpoint that uses a Pydantic AI agent to generate responses:

# file: my_django_agent/api.py

from django_bolt import BoltAPI
from django_bolt.responses import StreamingResponse
from pydantic_ai import Agent

api = BoltAPI()

agent = Agent(
    "anthropic:claude-haiku-4-5",
    instructions="Be concise, reply with one sentence.",
)


@api.get("/chat")
async def chat(prompt: str = "Where does 'hello world' come from?"):
    async def content():
        async with agent.run_stream(prompt) as result:
            async for chunk in result.stream_text(delta=True):
                yield chunk

    return StreamingResponse(content(), media_type="text/plain")

We use Claude here, but you can use any model supported by PydanticAI.

Test it out

Set ANTHROPIC_API_KEY in your environment:

export ANTHROPIC_API_KEY="your_anthropic_api_key_here"

Run the server:

uv run python manage.py runbolt --dev

Then curl the endpoint:

$ curl -N http://127.0.0.1:8000/chat

And watch the response stream in real time:

Or prompt it with a question:

$ curl -N -G --data-urlencode "prompt=what is Django" http://127.0.0.1:8000/chat

Django is a free, open-source Python web framework designed for building robust and scalable web applications quickly.%

Obviously this is a very simple example, but I think the simplicity can go a long way for people considering using Django for new LLM-related projects, since it’s already really easy with FastAPI. In fact, many examples on PydanticAI’s documentation, such as the frontend UI examples, use FastAPI, but it’s easy to do the same thing with Django-Bolt. And then you’ll have all of Django’s batteries included available, such as the admin interface, ORM, and unit testing framework. Options are nice to have!