FastStream is an asynchronous Python framework for building event-driven applications.

If you know FastAPI, you already know FastStream: the same decorators, type-driven validation, dependency injection and generated documentation — pointed at Kafka, RabbitMQ, NATS, Redis and MQTT instead of HTTP. It takes the boilerplate off your hands and leaves your broker intact.

FastStream simplifies the process of writing producers and consumers for message queues, handling all the parsing, lifecycle and documentation generation automatically.

Making streaming microservices has never been easier. The API is small enough to onboard a teammate in an afternoon, and it never costs you access to the broker underneath — approachable and complete are not a trade-off here. Here's a look at the core features that make FastStream a go-to framework for modern, data-centric microservices.

• A Spec You Never Write: a full AsyncAPI document generated from your handlers — the contract the neighbouring team keeps asking for, guaranteed to match the code, with an in-browser form for publishing test messages

• Tests Without a Broker: an in-memory test client runs your subscribers and publishers with validation intact — no containers in CI, no flakes, milliseconds instead of minutes

• Observable From Day One: OpenTelemetry traces, Prometheus metrics and Kubernetes probes come with the framework — a couple of middlewares instead of a few hundred lines in every service

• Your Broker, In Full: FastStream is a client for your broker, not a layer above all of them — Kafka consumer groups and partitioning, RabbitMQ exchanges and DLQ, NATS JetStream and KeyValue, Redis Streams, MQTT QoS. Five first-class clients that happen to share their ergonomics.

• Built-in Serialization: Leverage Pydantic or Msgspec validation capabilities to serialize and validate incoming messages

• Powerful Dependency Injection System: Manage your service dependencies efficiently with FastStream's built-in DI system

• Intuitive: Full-typed editor support makes your development experience smooth, catching errors before they reach runtime

• Extensible: Use extensions for lifespans, custom serialization and middleware

• Integrations: FastStream is fully compatible with any HTTP framework you want — including a dedicated FastAPI plugin, now shipped as its own package

That is FastStream: everything a messaging service needs around your handlers, and nothing between you and your broker.

Documentation: https://faststream.ag2.ai/latest/

• FastStream
  • Features
  • Versioning Policy
  • Installation
  • Writing app code
    • Pydantic serialization
    • Msgspec serialization
    • Your Broker, In Full
  • Testing the service
  • Running the application
  • Project Documentation
  • Dependencies
  • HTTP Frameworks integrations
    • Any Framework
    • FastAPI Plugin (deprecated)
  • Benchmarks
  • Stay in touch
  • Contributors

FastStream is a package based on the ideas and experiences gained from FastKafka and Propan. By joining our forces, we picked up the best from both packages and created a unified way to write services capable of processing streamed data regardless of the underlying protocol.

FastStream has a stable public API. Only major updates may introduce breaking changes.

Prior to FastStream's 1.0 release, each minor update is considered a major and can introduce breaking changes, but these changes were communicated through two-versions deprecation warnings prior to being fully removed. So features deprecated in the 0.4 version were only removed in version 0.6.

Our team is working toward the stable 1.0 version.

FastStream works on Linux, macOS, Windows and most Unix-style operating systems. You can install it with pip as usual:

pip install 'faststream[kafka]'
# or
pip install 'faststream[confluent]'
# or
pip install 'faststream[rabbit]'
# or
pip install 'faststream[nats]'
# or
pip install 'faststream[redis]'
# or
pip install 'faststream[mqtt]'

FastStream brokers provide convenient function decorators @broker.subscriber and @broker.publisher to allow you to delegate the actual process of:

• consuming and producing data to Event queues, and

• decoding and encoding JSON-encoded messages

These decorators make it easy to specify the processing logic for your consumers and producers, allowing you to focus on the core business logic of your application without worrying about the underlying integration.

Also, FastStream uses Pydantic to parse input JSON-encoded data into Python objects, making it easy to work with structured data in your applications, so you can serialize your input messages just using type annotations.

Here is an example Python app using FastStream that consumes data from an incoming data stream and outputs the data to another one:

from faststream import FastStream
from faststream.kafka import KafkaBroker
# from faststream.confluent import KafkaBroker
# from faststream.rabbit import RabbitBroker
# from faststream.nats import NatsBroker
# from faststream.redis import RedisBroker
# from faststream.mqtt import MQTTBroker

broker = KafkaBroker("localhost:9092")
# broker = RabbitBroker("amqp://guest:guest@localhost:5672/")
# broker = NatsBroker("nats://localhost:4222/")
# broker = RedisBroker("redis://localhost:6379/")
# broker = MQTTBroker("localhost")

app = FastStream(broker)

@broker.subscriber("in")
@broker.publisher("out")
async def handle_msg(user: str, user_id: int) -> str:
    return f"User: {user_id} - {user} registered"

Also, Pydantic’s BaseModel class allows you to define messages using a declarative syntax, making it easy to specify the fields and types of your messages.

from pydantic import BaseModel, Field, PositiveInt
from faststream import FastStream
from faststream.kafka import KafkaBroker

broker = KafkaBroker("localhost:9092")
app = FastStream(broker)

class User(BaseModel):
    user: str = Field(..., examples=["John"])
    user_id: PositiveInt = Field(..., examples=["1"])

@broker.subscriber("in")
@broker.publisher("out")
async def handle_msg(data: User) -> str:
    return f"User: {data.user} - {data.user_id} registered"

By default we use PydanticV2 written in Rust as serialization library, but you can downgrade it manually, if your platform has no Rust support - FastStream will work correctly with PydanticV1 as well.

To choose the Pydantic version, you can install the required one using the regular

pip install pydantic==1.X.Y

FastStream (and FastDepends inside) should work correctly with almost any version.

Moreover, FastStream is not tied to any specific serialization library, so you can use any preferred one. Fortunately, we provide a built‑in alternative for the most popular Pydantic replacement - Msgspec.

from fast_depends.msgspec import MsgSpecSerializer
from faststream.kafka import KafkaBroker

broker = KafkaBroker(serializer=MsgSpecSerializer())

You can read more about the feature in the documentation.

FastStream is a thin client, not an abstraction layer. It wraps your broker's own library — aiokafka or confluent-kafka, aio-pika, nats-py, redis-py, zmqtt — and takes over what every service otherwise rewrites by hand: lifecycle, serialization, acknowledgement, observability, documentation, tests. What the broker itself offers stays yours.

Two rules follow, and they explain most of our API decisions:

1. We do not implement business logic. No retries, no delayed delivery, no task orchestration. Those are architectural choices, and a framework that makes them for you owns your architecture.
2. Every native broker feature stays reachable. When the ergonomic path is not enough, the client underneath is one annotation away — Connection and Channel for RabbitMQ, Consumer for Kafka, Client for NATS and MQTT, Redis for Redis. Every broker we support has one.

What the five clients share is a deliberately small surface:

from faststream.[broker] import [Broker], [Broker]Message

broker = [Broker](*servers)

@broker.subscriber([source])  # Kafka topic / RMQ queue / NATS subject / MQTT topic / etc
@broker.publisher([destination])  # topic / routing key / subject / etc
async def handler(msg: [Broker]Message) -> None:
    await msg.ack()  # control brokers' acknowledgement policy

...

await broker.publish("Message", [destiination])

Beyond this scope you can use any broker-native features you need:

• Kafka - specific partition reads, partitioner control, consumer groups, batch processing, etc.
• RabbitMQ - all exchange types, Redis Streams, RPC, manual channel configuration, DLQ, etc.
• NATS - core and Push/Pull JetStream subscribers, KeyValue, ObjectStorage, RPC, etc.
• Redis - Pub/Sub, List, Stream subscribers, consumer groups, acknowledgements, etc.
• MQTT - topic subscriptions (including wildcards), QoS and retain, MQTT 3.1.1 and 5.0, request/reply (RPC), TLS, etc.

You can find detailed information about all supported features in FastStream’s broker‑specific documentation.

If a particular feature is missing or not yet supported, you can always fall back to the native broker client/connection for those operations.

The service can be tested using the TestBroker context managers, which, by default, puts the Broker into "testing mode".

The Tester will redirect your subscriber and publisher decorated functions to the InMemory brokers, allowing you to quickly test your app without the need for a running broker and all its dependencies.

Using pytest, the test for our service would look like this:

import pytest
import pydantic
from faststream.kafka import TestKafkaBroker


@pytest.mark.asyncio
async def test_correct():
    async with TestKafkaBroker(broker) as br:
        await br.publish({
            "user": "John",
            "user_id": 1,
        }, "in")

@pytest.mark.asyncio
async def test_invalid():
    async with TestKafkaBroker(broker) as br:
        with pytest.raises(pydantic.ValidationError):
            await br.publish("wrong message", "in")

The application can be started using the built-in FastStream CLI command.

Before running the service, install FastStream CLI using the following command:

pip install "faststream[cli]"

To run the service, use the FastStream CLI command and pass the module (in this case, the file where the app implementation is located) and the app symbol to the command.

faststream run basic:app

After running the command, you should see the following output:

INFO     - FastStream app starting...
INFO     - input_data |            - `HandleMsg` waiting for messages
INFO     - FastStream app started successfully! To exit press CTRL+C

Also, FastStream provides you with a great hot reload feature to improve your Development Experience

faststream run basic:app --reload

And multiprocessing horizontal scaling feature as well:

faststream run basic:app --workers 3

You can learn more about CLI features here

FastStream automatically generates documentation for your project according to the AsyncAPI specification. You can work with both generated artifacts and place a web view of your documentation on resources available to related teams.

The availability of such documentation significantly simplifies the integration of services: you can immediately see what channels and message formats the application works with. And most importantly, it won't cost anything - FastStream has already created the docs for you!

FastStream (thanks to FastDepends) has a dependency management system similar to pytest fixtures and FastAPI Depends at the same time. Function arguments declare which dependencies are needed, and a special decorator delivers them from the global Context object.

from typing import Annotated
from faststream import Depends, Logger

async def base_dep(user_id: int) -> bool:
    return True

@broker.subscriber("in-test")
async def base_handler(user: str,
                       logger: Logger,
                       dep: Annotated[bool, Depends(base_dep)]):
    assert dep is True
    logger.info(user)

You can use FastStream MQBrokers without a FastStream application. Just start and stop them according to your application's lifespan.

from aiohttp import web

from faststream.kafka import KafkaBroker

broker = KafkaBroker("localhost:9092")

@broker.subscriber("test")
async def base_handler(body):
    print(body)

async def start_broker(app):
    await broker.start()

async def stop_broker(app):
    await broker.stop()

async def hello(request):
    return web.Response(text="Hello, world")

app = web.Application()
app.add_routes([web.get("/", hello)])
app.on_startup.append(start_broker)
app.on_cleanup.append(stop_broker)

if __name__ == "__main__":
    web.run_app(app)

Deprecated. The integration has been moved to the faststream_fastapi package and will be removed in the 1.0.0 version:

pip install faststream_fastapi

Also, FastStream can be used as part of FastAPI.

Just import a StreamRouter you need and declare the message handler with the same @router.subscriber(...) and @router.publisher(...) decorators.

from fastapi import FastAPI
from pydantic import BaseModel

from faststream.kafka.fastapi import KafkaRouter

router = KafkaRouter("localhost:9092")

class Incoming(BaseModel):
    m: dict

@router.subscriber("test")
@router.publisher("response")
async def hello(m: Incoming):
    return {"response": "Hello, world!"}

app = FastAPI()
app.include_router(router)

More integration features can be found here

We use codspeed to run benchmarks for both FastStream itself and raw clients.

FastStream is used by research institutions, public sector organizations and companies — among them ECMWF, Hydro-Québec, the Rubin Observatory, NERSC and Red Hat. Neighbouring projects such as Pydantic Logfire, RabbitMQ and EMQX maintain a FastStream integration of their own.

See the full list on the Used By page, and open a pull request to add your own project.

Please show your support and stay in touch by:

• giving our GitHub repository a star, and

• joining the discussions on GitHub

• joining our Telegram group

Your support helps us to stay in touch with you and encourages us to continue developing and improving the framework. Thank you for your support!

Thanks to all of these amazing people who made the project better!