Starting with Streamlit 1.53, we're introducing experimental support for running Streamlit with Starlette as the underlying web framework. This is the first step toward our long-term goal of fully migrating from Tornado to Starlette/ASGI, and we need your help to identify any issues or gaps before the full transition.

What's New

Note: To use the experimental Starlette features, you need to install the optional dependencies:

pip install streamlit[starlette]

1. Starlette Server Mode (Experimental)

You can now run Streamlit using Starlette instead of Tornado by setting a config option:

# .streamlit/config.toml
[server]
useStarlette = true

Or via command line:

streamlit run app.py --server.useStarlette=true

2. ASGI App Entry Point (Experimental)

For advanced use cases, you can now create an ASGI-compatible Streamlit application using the new App class:

from streamlit.starlette import App

app = App("main.py")

Run it with streamlit run (preferred):

streamlit run myapp.py

Streamlit will automatically detect the App instance and run it in Starlette mode. Alternatively, you can use any ASGI server directly:

uvicorn myapp:app --host 0.0.0.0 --port 8501

The App class supports advanced configuration including:

• Custom HTTP routes — Add REST API endpoints alongside your Streamlit app
• Middleware — Add security headers, authentication, logging, etc.
• Lifespan hooks — Run code on startup/shutdown (e.g., pre-warm caches, initialize resources)
• Framework integration — Mount Streamlit inside FastAPI, Django, or other ASGI frameworks

📖 For more details on the App interface and its full capabilities, see the specification in PR #13449.

from streamlit.starlette import App
from starlette.routing import Route
from starlette.responses import JSONResponse

async def health_check(request):
    return JSONResponse({"status": "healthy", "service": "streamlit-app"})

async def get_data(request):
    return JSONResponse({"items": ["apple", "banana", "cherry"], "count": 3})

app = App(
    "main.py",
    routes=[
        Route("/api/health", health_check),
        Route("/api/data", get_data),
    ],
)

from streamlit.starlette import App
from starlette.middleware import Middleware
from starlette.middleware.base import BaseHTTPMiddleware

class SecurityHeadersMiddleware(BaseHTTPMiddleware):
    async def dispatch(self, request, call_next):
        response = await call_next(request)
        response.headers["X-Content-Type-Options"] = "nosniff"
        response.headers["X-Frame-Options"] = "SAMEORIGIN"
        return response

app = App(
    "main.py",
    middleware=[Middleware(SecurityHeadersMiddleware)],
)

from fastapi import FastAPI
from starlette.routing import Mount
from streamlit.starlette import App

# Create FastAPI sub-application
api = FastAPI()

@api.get("/health")
async def health():
    return {"status": "healthy"}

@api.post("/predict")
async def predict(data: dict):
    return {"prediction": data.get("value", 0) * 2}

# Mount FastAPI into Streamlit app
# - Streamlit UI at /
# - FastAPI endpoints at /api/*
# - FastAPI docs at /api/docs
app = App(
    "dashboard.py",
    routes=[Mount("/api", app=api)],
)

from fastapi import FastAPI
from streamlit.starlette import App

# Create the Streamlit sub-application
streamlit_app = App("dashboard.py")

# Create FastAPI with Streamlit's lifespan to manage the runtime lifecycle
api = FastAPI(lifespan=streamlit_app.lifespan())

@api.get("/api/data")
async def get_data():
    return {"data": [1, 2, 3]}

# Mount Streamlit under /dashboard
api.mount("/dashboard", streamlit_app)

# Run with: uvicorn myapp:api

from contextlib import asynccontextmanager
from streamlit.starlette import App

@asynccontextmanager
async def lifespan(app):
    # Startup: runs before accepting connections
    print("🚀 Starting up...")
    model = load_ml_model()  # Pre-warm caches
    db = init_db_connection()

    # Yield state accessible via app.state
    yield {"model": model, "db": db}

    # Shutdown: cleanup resources
    print("👋 Shutting down...")
    db.close()

app = App("main.py", lifespan=lifespan)

Why We're Migrating to Starlette/ASGI

The migration to Starlette/ASGI will unlock numerous frequently-requested features:

Custom HTTP Endpoints

• Support custom HTTP requests #439
• Allow streamlit apps to easily serve robots.txt at the root level #9673
• Widget to host folder as website (for preview) #6195
• Enhanced JavaScript File Serving in Streamlit #9090
• MCP server for streamlit #11333
• Custom 404 / Error Page / Maintenance Page that is displayed whenever a bug is pushed or the app encounters and issue on boot #8713

Security Headers & Middleware

• Handling Security Headers in Streamlit #6417
• Allow configuring HTTP Security Headers #9160
• Ability to write cookies #861
• Support ip whitelist #8823

Framework Integration & ASGI Support

• Provide way to run with the WSGI/ASGI protocols #4311
• Integration with Flask - render simple Streamlit component on Flask or Fast API #4567
• Enable Streamlit to run multiple applications on Django via Django Channels instead of Tornado #927
• Multiple Runtime instances #7546

Server Configuration & Lifecycle

• On session start/shutdown hooks #7688
• Add a user entrypoint executed on startup before page loding #8991
• Pre-cache on start-up #6108
• Tornado HTTPServer extra arguments #9916
• Shut down the session after closing the browser tab by the user #8545
• Allow user using setup script to interact with streamlit app out of Script runner #7667

SEO & Metadata

• Allow specifying custom website metadata #5673
• Ability to embed metadata for SEO indexing with Google Search Console, Analytics or Tag Manager trackers #8999

We Need Your Feedback

Since we plan to eventually make Starlette the default (and only) server backend, we want to catch any compatibility issues early. Please help us by:

1. Try it out

Enable Starlette mode in your existing apps:

streamlit run your_app.py --server.useStarlette=true

2. Report any issues

If you encounter problems, please comment below with:

• What happened — Error messages, unexpected behavior
• Your environment — OS, Python version, Streamlit version
• How to reproduce — Minimal code example if possible
• Comparison — Does the issue occur with the default Tornado server?

3. Share your use cases

We're especially interested in:

• ⚡ Performance comparisons (response times, memory usage, concurrent users)
• 🔌 Integration scenarios (FastAPI, Django, other ASGI frameworks)
• 🛡️ Security/middleware requirements
• 🚀 Deployment configurations (uvicorn, gunicorn, hypercorn, cloud platforms)

Known Limitations

• This is experimental — APIs may change before the final release
• TBD

Thank you for helping us build a better Streamlit! Your early feedback is invaluable in ensuring a smooth transition for the entire community. 🙏

Please comment below with your experiences, questions, or concerns.