Web development is one of the most popular use cases for programming. Python is one of the most popular programming languages in the world. So why can't we build web apps in Python? Making a UI should be simple, but even you have great engineers on your team, the overhead of learning a new language and tools was a huge barrier. Often making a UI could be harder than the actual work one is doing! We built Reflex, an open source Python web framework to solve this exact problem. TLDR Under the hood, Reflex apps compile down to a React frontend app and a FastAPI backend app. Only the UI is compiled to Javascript; all the app logic and state management stays in Python and is run on the server. Reflex uses WebSockets to send events from the frontend to the backend, and to send state updates from the backend to the frontend. Existing Python Solutions There were a few ways already to build apps in Python, but none of them fit our needs. On the one hand, there are frameworks like Django and Flask that are great for building production-grade web apps. But they only handle the backend - you still need to use JavaScript and a frontend framework, as well as writing a lot of boilerplate code to connect the frontend and backend. On the other hand, pure Python libraries like Dash and Streamlit can be great for small projects, but they are limited to a specific use case and don't have the features and performance to build a full web app. As your app grows in features and complexity, you may find yourself hitting the limits of the framework, at which point you either have to limit your idea to fit the framework, or scrap your project and rebuild it using a "real web framework". We want to bridge this gap by creating a framework that is easy and intuitive to get started with, while remaining flexible and powerful to support any app. Goals of Reflex Pure Python: Use one language for everything.
Easy to get started: Build your ideas easily without needing web development experience.
Full flexibility: Web apps should match the customizability and performance of traditional web frameworks.
Batteries included: Handle the full-stack from the frontend, to the backend, to deployment. Now let's dive into how we built Reflex to meet these goals. The Reflex Architecture Full-stack web apps are made up of a frontend and a backend. The frontend is the user interface, and is served as a web page that runs on the user's browser. The backend handles the logic and state management (such as databases and APIs), and is run on a server. In traditional web development, these are usually two separate apps, and are often written in different frameworks or languages. For example, you may combine a Flask backend with a React frontend. With this approach, you have to maintain two separate apps and end up writing a lot of boilerplate code to connect the frontend and backend. We want to simplify this process in Reflex by defining both the frontend and backend in a single codebase, while using Python for everything. Developers should only worry about their app's logic and not about the low-level implementation details. Frontend We want Reflex apps to look and feel like a traditional web app to the end user, while still being easy to build and maintain for the developer. To do this, we built on top of mature and popular web technologies. When you reflex run your app, Reflex compiles the frontend down to a single-page Next.js app and serves it on a port (by default 3000) that you can access in your browser. The frontend's job is to reflect the app's state, and send events to the backend when the user interacts with the UI. No actual logic is run on the frontend. Components Reflex frontends are built using components that can be composed together to create complex UIs. Instead of using a templating language that mixes HTML and Python, we just use Python functions to define the UI. Under the hood, components compile down to React components. Many of our core components are based on Radix, a popular React component library. We also have many other components for graphing, datatables, and more. We chose React because it is a popular library with a huge ecosystem. Our goal isn't to recreate the web ecosystem, but to make it accessible to Python developers. This also lets our users bring their own components if we don't have a component they need. Users can wrap their own React components and then publish them for others to use. Over time we will build out our third party component ecosystem so that users can easily find and use components that others have built. Styling We wanted to make sure Reflex apps look good out of the box, while still giving developers full control over the appearance of their app. We have a core theming system that lets you set high level styling options such as dark mode and accent color throughout your app to give it a unified look and feel. Beyond this, Reflex components can be styled using the full power of CSS. We leverage the Emotion library to allow "CSS-in-Python" styling, so you can pass any CSS prop as a keyword argument to a component. This includes responsive props by passing a list of values. Backend In Reflex only the frontend compiles to Javascript and runs on the user's browser, while all the state and logic stays in Python and is run on the server. When you reflex run, we start a FastAPI server (by default on port 8000) that the frontend connects to through a websocket. All the state and logic are defined within a State class. The state is made up of vars and event handlers. Vars are any values in your app that can change over time. They are defined as class attributes on your State class, and may be any Python type that can be serialized to JSON. Event handlers are methods in your State class that are called when the user interacts with the UI. They are the only way that we can modify the vars in Reflex, and can be called in response to user actions, such as clicking a button or typing in a text box. Since event handlers are run on the backend, you can use any Python library within them. Event Processing Normally when writing web apps, you have to write a lot of boilerplate code to connect the frontend and backend. With Reflex, you don't have to worry about that - we handle the communication between the frontend and backend for you. Developers just have to write their event handler logic, and when the vars are updated the UI is automatically updated. Event Triggers The user can interact with the UI in many ways, such as clicking a button, typing in a text box, or hovering over an element. In Reflex, we call these event triggers. Event Queue On the frontend, we maintain an event queue of all pending events. An event consists of three major pieces of data: client token: Each client (browser tab) has a unique token to identify it. This let's the backend know which state to update.
event handler: The event handler to run on the state.
arguments: The arguments to pass to the event handler. When an event is triggered, it is added to the queue. We have a processing flag to make sure only one event is processed at a time. This ensures that the state is always consistent and there aren't any race conditions with two event handlers modifying the state at the same time. There are exceptions to this, such as background events which allow you to run events in the background without blocking the UI. Once the event is ready to be processed, it is sent to the backend through a WebSocket connection. State Manager Once the event is received, it is processed on the backend. Reflex uses a state manager which maintains a mapping between client tokens and their state. By default, the state manager is just an in-memory dictionary, but it can be extended to use a database or cache. In production we use Redis as our state manager. Event Handling Once we have the user's state, the next step is to run the event handler with the arguments. State Updates Every time an event handler returns (or yields), we save the state in the state manager and send the state updates to the frontend to update the UI. To maintain performance as your state grows, internally Reflex keeps track of vars that were updated during the event handler (dirty vars). When the event handler is done processing, we find all the dirty vars and create a state update to send to the frontend. We store the new state in our state manager, and then send the state update to the frontend. The frontend then updates the UI to reflect the new state. Conclusion I hope this provides a good overview of how Reflex works under the hood. We will have more posts coming out to share how we made Reflex scalable and performant through features such as state sharding and compiler optimizations. Web development is one of the most popular use cases for programming. Python is one of the most popular programming languages in the world. So why can't we build web apps in Python? Making a UI should be simple, but even you have great engineers on your team, the overhead of learning a new language and tools was a huge barrier. Often making a UI could be harder than the actual work one is doing! We built Reflex, an open source Python web framework to solve this exact problem. TLDR Under the hood, Reflex apps compile down to a React frontend app and a FastAPI backend app. Only the UI is compiled to Javascript; all the app logic and state management stays in Python and is run on the server. Reflex uses WebSockets to send events from the frontend to the backend, and to send state updates from the backend to the frontend. React FastAPI WebSockets Existing Python Solutions There were a few ways already to build apps in Python, but none of them fit our needs. On the one hand, there are frameworks like Django and Flask that are great for building production-grade web apps. But they only handle the backend - you still need to use JavaScript and a frontend framework, as well as writing a lot of boilerplate code to connect the frontend and backend. Django Flask On the other hand, pure Python libraries like Dash and Streamlit can be great for small projects, but they are limited to a specific use case and don't have the features and performance to build a full web app. As your app grows in features and complexity, you may find yourself hitting the limits of the framework, at which point you either have to limit your idea to fit the framework, or scrap your project and rebuild it using a "real web framework". Dash Streamlit We want to bridge this gap by creating a framework that is easy and intuitive to get started with, while remaining flexible and powerful to support any app. Goals of Reflex Pure Python: Use one language for everything. Easy to get started: Build your ideas easily without needing web development experience. Full flexibility: Web apps should match the customizability and performance of traditional web frameworks. Batteries included: Handle the full-stack from the frontend, to the backend, to deployment. Pure Python : Use one language for everything. Pure Python Easy to get started : Build your ideas easily without needing web development experience. Easy to get started Full flexibility : Web apps should match the customizability and performance of traditional web frameworks. Full flexibility Batteries included : Handle the full-stack from the frontend, to the backend, to deployment. Batteries included Now let's dive into how we built Reflex to meet these goals. The Reflex Architecture Full-stack web apps are made up of a frontend and a backend. The frontend is the user interface, and is served as a web page that runs on the user's browser. The backend handles the logic and state management (such as databases and APIs), and is run on a server. In traditional web development, these are usually two separate apps, and are often written in different frameworks or languages. For example, you may combine a Flask backend with a React frontend. With this approach, you have to maintain two separate apps and end up writing a lot of boilerplate code to connect the frontend and backend. We want to simplify this process in Reflex by defining both the frontend and backend in a single codebase, while using Python for everything. Developers should only worry about their app's logic and not about the low-level implementation details. Frontend We want Reflex apps to look and feel like a traditional web app to the end user, while still being easy to build and maintain for the developer. To do this, we built on top of mature and popular web technologies. When you reflex run your app, Reflex compiles the frontend down to a single-page Next.js app and serves it on a port (by default 3000 ) that you can access in your browser. reflex run Next.js 3000 The frontend's job is to reflect the app's state, and send events to the backend when the user interacts with the UI. No actual logic is run on the frontend. Components Reflex frontends are built using components that can be composed together to create complex UIs. Instead of using a templating language that mixes HTML and Python, we just use Python functions to define the UI. Under the hood, components compile down to React components. Many of our core components are based on Radix , a popular React component library. We also have many other components for graphing, datatables, and more. We chose React because it is a popular library with a huge ecosystem. Our goal isn't to recreate the web ecosystem, but to make it accessible to Python developers. Radix This also lets our users bring their own components if we don't have a component they need. Users can wrap their own React components and then publish them for others to use. Over time we will build out our third party component ecosystem so that users can easily find and use components that others have built. Styling We wanted to make sure Reflex apps look good out of the box, while still giving developers full control over the appearance of their app. We have a core theming system that lets you set high level styling options such as dark mode and accent color throughout your app to give it a unified look and feel. Beyond this, Reflex components can be styled using the full power of CSS. We leverage the Emotion library to allow "CSS-in-Python" styling, so you can pass any CSS prop as a keyword argument to a component. This includes responsive props by passing a list of values. Emotion Backend In Reflex only the frontend compiles to Javascript and runs on the user's browser, while all the state and logic stays in Python and is run on the server. When you reflex run , we start a FastAPI server (by default on port 8000 ) that the frontend connects to through a websocket. reflex run 8000 All the state and logic are defined within a State class. The state is made up of vars and event handlers . Vars are any values in your app that can change over time. They are defined as class attributes on your State class, and may be any Python type that can be serialized to JSON. State vars event handlers State Event handlers are methods in your State class that are called when the user interacts with the UI. They are the only way that we can modify the vars in Reflex, and can be called in response to user actions, such as clicking a button or typing in a text box. State Since event handlers are run on the backend, you can use any Python library within them. Event Processing Normally when writing web apps, you have to write a lot of boilerplate code to connect the frontend and backend. With Reflex, you don't have to worry about that - we handle the communication between the frontend and backend for you. Developers just have to write their event handler logic, and when the vars are updated the UI is automatically updated. Event Triggers The user can interact with the UI in many ways, such as clicking a button, typing in a text box, or hovering over an element. In Reflex, we call these event triggers . event triggers Event Queue On the frontend, we maintain an event queue of all pending events. An event consists of three major pieces of data: client token: Each client (browser tab) has a unique token to identify it. This let's the backend know which state to update. event handler: The event handler to run on the state. arguments: The arguments to pass to the event handler. client token : Each client (browser tab) has a unique token to identify it. This let's the backend know which state to update. client token event handler : The event handler to run on the state. event handler arguments : The arguments to pass to the event handler. arguments When an event is triggered, it is added to the queue. We have a processing flag to make sure only one event is processed at a time. This ensures that the state is always consistent and there aren't any race conditions with two event handlers modifying the state at the same time. There are exceptions to this, such as background events which allow you to run events in the background without blocking the UI. processing Once the event is ready to be processed, it is sent to the backend through a WebSocket connection. State Manager Once the event is received, it is processed on the backend. Reflex uses a state manager which maintains a mapping between client tokens and their state. By default, the state manager is just an in-memory dictionary, but it can be extended to use a database or cache. In production we use Redis as our state manager. state manager Event Handling Once we have the user's state, the next step is to run the event handler with the arguments. State Updates Every time an event handler returns (or yields), we save the state in the state manager and send the state updates to the frontend to update the UI. To maintain performance as your state grows, internally Reflex keeps track of vars that were updated during the event handler ( dirty vars ). state updates dirty vars When the event handler is done processing, we find all the dirty vars and create a state update to send to the frontend. We store the new state in our state manager, and then send the state update to the frontend. The frontend then updates the UI to reflect the new state. Conclusion I hope this provides a good overview of how Reflex works under the hood. We will have more posts coming out to share how we made Reflex scalable and performant through features such as state sharding and compiler optimizations.

This post provides insights into new product. 

Designing a Pure Python Web Framework

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

Untitled Story

$DEFI Token Hits 7 Major Exchanges: A Milestone Achievement

$JTC Network To List On BitMart Exchange

$500k Presale: TG.Casino Passes Milestone with Upcoming Telegram-Powered Platform

$3 Million in Seed Funding for Web3 Founders Announced By Necto Labs

$2M Backing and a Vision: How GAM3S.GG is Reshaping Web3 Gaming

$DEFI Token Hits 7 Major Exchanges: A Milestone Achievement

$JTC Network To List On BitMart Exchange

$500k Presale: TG.Casino Passes Milestone with Upcoming Telegram-Powered Platform

$3 Million in Seed Funding for Web3 Founders Announced By Necto Labs

$2M Backing and a Vision: How GAM3S.GG is Reshaping Web3 Gaming

Light-Mode

Classic

Newspaper

Dark-Mode

Neon Noir

Minty

HN StartUps