The fastest Python web framework — built on Django

Rust-powered HTTP, msgspec serialization, full type validation —
with the Django ORM, Django Admin, and every Django package you already use.

Documentation · Quick Start · Features · Benchmarks · Video Tutorial · Discord

Django-Bolt is the fastest Python web framework: 300k+ requests/second on a single 12-core desktop (8 processes, C=100, loopback), ahead of FastAPI and Robyn, and even of Bun-based JavaScript frameworks (Elysia, Hono) on JSON payloads. It is a fully typed API framework for Django. It serves your endpoints from a Rust HTTP server (Actix Web + Tokio), bridges to your Python handlers with PyO3, and serializes with msgspec — while everything you love about Django (ORM, Admin, auth, middleware, signals, third-party apps) keeps working out of the box.

Think Django REST Framework or Django Ninja, with a Rust engine underneath and no gunicorn or uvicorn required.

from django_bolt import BoltAPI

api = BoltAPI()

@api.get("/hello/{name}")
async def hello(name: str):
    return {"message": f"Hello, {name}!"}

python manage.py runbolt --dev

✨ Why Django-Bolt?

🚀 Quick Start

1. Install

pip install django-bolt      # or: uv add django-bolt

2. Add to INSTALLED_APPS

# myproject/settings.py
INSTALLED_APPS = [
    ...,
    "django_bolt",
]

3. Write your first endpoint

Create an api.py next to your settings.py (or inside any Django app — Bolt autodiscovers them all):

# myproject/api.py
import msgspec
from django.contrib.auth import get_user_model
from django_bolt import BoltAPI

User = get_user_model()
api = BoltAPI()


class UserSchema(msgspec.Struct):
    id: int
    username: str


@api.get("/users/{user_id}")
async def get_user(user_id: int) -> UserSchema:   # response is type-validated
    user = await User.objects.aget(id=user_id)    # Django ORM, no extra setup
    return {"id": user.id, "username": user.username}

4. Run

python manage.py runbolt --dev              # auto-reload for development
python manage.py runbolt --processes 4      # production: multi-process, no gunicorn/uvicorn

Your API is live at http://localhost:8000/users/1 and interactive docs at http://localhost:8000/docs.

📖 Next: the Quick Start guide → Deployment → Topic guides.

🧭 A tour of the API

import msgspec
from typing import Annotated
from django_bolt import BoltAPI
from django_bolt.param_functions import Header

api = BoltAPI()

class CreateUser(msgspec.Struct):
    username: str
    email: str

@api.post("/users", status_code=201)
async def create_user(
    user: CreateUser,                                     # JSON body → validated struct
    api_key: Annotated[str, Header("x-api-key")],         # header
    page: int = 1,                                        # query param with default
):
    return {"username": user.username, "page": page}

from django_bolt.auth import JWTAuthentication, IsAuthenticated, Requires

IsStaff = Requires("is_staff", True)

@api.get("/admin/stats", auth=[JWTAuthentication()], guards=[IsAuthenticated(), IsStaff])
async def admin_stats(request):
    return {"user_id": request.user.id}

from django_bolt import ModelViewSet, PageNumberPagination
from django_bolt.serializers import Serializer
from myapp.models import Article

class ArticleSchema(Serializer):
    id: int
    title: str
    content: str

    class Config:
        field_sets = {"list": ["id", "title"]}

@api.viewset("/articles")
class ArticleViewSet(ModelViewSet):
    queryset = Article.objects.all()
    serializer_class = ArticleSchema
    pagination_class = PageNumberPagination

from django_bolt import WebSocket, StreamingResponse

@api.websocket("/ws/echo")
async def echo(websocket: WebSocket):
    await websocket.accept()
    async for message in websocket.iter_text():
        await websocket.send_text(f"Echo: {message}")

@api.get("/events")
async def events():
    async def stream():
        for i in range(10):
            yield f"data: tick {i}\n\n"
    return StreamingResponse(stream(), media_type="text/event-stream")

from bolt_mcp import MCP          # pip install "django-bolt[mcp]"

mcp = MCP("my-server")

@mcp.tool
async def add(a: int, b: int) -> dict:
    return {"sum": a + b}

api.mount_mcp(mcp)

from django_bolt.middleware import cors, rate_limit

@api.get("/public")
@cors(origins=["https://example.com"])
@rate_limit(rps=100, burst=200)
async def public():
    return {"ok": True}

📦 Features

All runtime settings and environment variables are listed in the Settings reference.

📊 Benchmarks

Measured with bombardier on a single 12-core desktop (Ryzen 5 5600G), loopback, C=100, N=100000, 8 processes × 1 worker (runbolt --processes 8). Absolute numbers are hardware-specific; run just save-bench to reproduce on your machine. Full results: python/benchmark/BENCHMARK.md.

Server-Sent Events, 10,000 concurrent clients for 60 s: 9,489 msg/s, 100% connections succeeded, ~236 MB RSS, 11.9% average CPU.

Against JavaScript runtimes

The same JSON payloads served by Django-Bolt, Elysia (Bun), and Hono (Bun & Node), 8 processes each:

Why so fast?

• Actix Web + Tokio handle HTTP parsing and responses; matchit routes with zero-copy path matching.
• Auth, guards, CORS, rate limiting, compression run in Rust — no GIL, no Python per-request overhead.
• msgspec serialization is 5–10× faster than the standard library; response bodies cross to Rust zero-copy.
• Sync-dispatch bypass: handlers that don't actually await are detected at registration and skip the async bridge entirely.
• Registration-time precomputation: parameter extraction, dependency graphs, and middleware are compiled once, reused forever.

🏗️ How it works

HTTP request
   │
   ▼
Actix Web (Rust) ── routing (matchit) ── CORS · rate limit · compression
   │
   ▼
Auth & guards (Rust, no GIL) ── JWT / API key / session · IsAuthenticated · Requires(...)
   │
   ▼
Dispatch ── sync fast path (single GIL block)  or  async path (persistent worker loop)
   │
   ▼
Your handler ── typed params · Depends(...) · Django ORM
   │
   ▼
msgspec serialization ── zero-copy body ── HTTP response

🚢 Deployment

python manage.py runbolt --host 0.0.0.0 --port 8000 --processes 4
python manage.py runbolt --processes 4 --max-rss 512   # recycle workers above 512 MB

Multi-process scaling uses SO_REUSEPORT for kernel-level load balancing. Worker recycling, crash respawn, graceful shutdown, and WebSocket drain are built in. See the Deployment guide for systemd, supervisor, and reverse-proxy setups.

🤝 Contributing

Contributions are welcome! See CONTRIBUTING.md for the development setup (Rust toolchain, uv, just), the test workflow, and pull request guidelines.

git clone https://github.com/dj-bolt/django-bolt.git && cd django-bolt
uv sync && just build && just test-py

💬 Community

• 📖 Documentation
• 💬 Discord
• 🐛 Issues
• 🎥 Video walkthrough by BugBytes
• 🤖 Ask DeepWiki · for AI assistants: llms.txt
• ❓ FAQ · Comparison vs Ninja / DRF / FastAPI / Litestar · How it works

💖 Sponsors

Support Django-Bolt's development by becoming a sponsor. Your logo will appear here with a link to your website.

Backers

🙏 Acknowledgments

Django-Bolt stands on the shoulders of giants:

• Django REST Framework — ViewSet patterns, permission system, and overall API philosophy
• FastAPI — dependency injection, parameter extraction, and type-hint-driven design
• Litestar — OpenAPI plugin architecture, middleware and guard design
• Robyn — proved the potential of Rust-powered Python web frameworks with PyO3
• Actix Web, PyO3, msgspec, matchit — the foundations that make the speed possible

📄 License

Django-Bolt is released under the MIT License.