What Can You Build in Flet?!

追加: 2026-05-25

[00:00:00]7 million PyPI downloads, 16,000 GitHub stars, an API that has been 99% stable for 6 months, and as of May 2026, no firm release date for version 1.0. It's targeted for 2026, but no month is committed. That's Flet, a Python framework that with one file ships an app to iOS, Android, Windows, macOS, Linux, and the web. Almost no other Python framework does that cleanly.

[00:00:26]There's also no other Python framework whose existence is structurally a confession. A confession that Python, the language that conquered data science, machine learning, web backends, infrastructure automation, and most recently its own developer tooling, has one frontier it cannot own cleanly, the screen. Three things. What Flet actually is mechanically, because it's not what it looks like in the marketing. What you can ship with it right now, in May 2026, before 1.0 lands. And the structural thing that every Python UI framework, Flet, Reflex, Streamlit, NiceGUI, all of them, is quietly leaving off the landing page. Stay till the end and I'll tell you which proxy contract each one is asking you to sign. Picking the wrong one is how Python projects ship a demo on Friday and ship nothing 6 months later. What Flet actually is. You write Python. Flet takes that Python and builds a tree of Flutter widgets, Material and Cupertino, the same primitives Google ships in the Android and iOS Flutter SDKs. Flutter renders.

[00:01:25]The rendering happens in Skia, the C++ graphics engine that Chrome also uses.

[00:01:30]So, when you write Flet, you're sending instructions across three layers. Python at the top, a web socket or compiled bridge in the middle, Flutter and Skia at the bottom. Your Python code doesn't draw anything. It tells something else to draw. The same Python file targets six surfaces. Run it as a desktop app and Flet bundles a Flutter runtime into a standalone executable.

[00:01:53]app and Flet serves a Flutter web build.

[00:01:55]Run Flet build for iOS or Android and you get a real binary that ships to the App Store or Play Store. Flet's own dev tool app did exactly that, written in Python, packaged with Flet build, accepted by Apple review. That was the proof. There's a fourth mode that gets less attention. As of version 0.4, Flet apps can be packaged as a standalone static website using Pyodide, Python compiled to WebAssembly. No server, no back end, just a folder you drop on GitHub pages or Cloudflare pages. Python running in the browser. Single thread, but it works. The current release is 0.85.1, shipped in May 2026. It moved Flet's new declarative component model from beta to production ready, what you can ship right now. Five categories. One, internal tools and dashboards, the sweet spot. If you have a Tkinter app from 2018 or a Streamlit dashboard that has outgrown its rerun model, this is the direct upgrade. Material design, native widgets, runs everywhere your team installs it. Two, cross-platform desktop apps. The Flet pack command produces a standalone executable for Windows, macOS, or Linux. No installer fight, no separate code bases. Three, mobile MVPs.

[00:03:05]This is the unique offer. Among web-first Python frameworks, there's no faster path from a single code base to both app stores. Beeware takes a different route, native OS widgets, the Toga toolkit, but that's a different trade and a different ecosystem. Whether the app should ultimately stay in Flet is a separate question. For validating an idea, it's untouched. Four, standalone web SPAs with no back end, static Pyodide deployment. You can put a real Python application with Pandas, with NumPy, on GitHub pages with nothing running on a server. For calculators, interactive tutorials, internal tools that compliance won't let you host server-side, this is real. Five, data-heavy apps where Streamlit is too restrictive. Streamlit's top-down rerun model is great until your app needs persistent state, multi-step forms, or a real navigation stack, Flet gives you those. What you should not ship in Flet in 2026, consumer apps where bundle size matters on the first download, games, anything where the rendering is the product rather than the interface, anything that needs rich text editing or real-time video. Those are documented limitations, not minor ones. The competition, Streamlit, best in class for data dashboards, top-down script model, every interaction reruns the entire script, brilliant for analysts, painful for stateful apps. Reflex, Python on top, but it compiles your UI to React and your backend to FastAPI, web only. If your target is a web app and nothing else, Reflex is the most React-shaped option in Python. NiceGUI, FastAPI plus a Vue-based UI layer, built for monitoring tools, admin panels, things that live behind your firewall.

[00:04:46]Flet, the only one of these four that ships true mobile binaries from a Python code base. Flutter-rendered, cross-platform from a single file. Also the only one where the rendering layer is Flutter, not React. The choice isn't about which is best, it's about which trade you can live with. Here's the part none of the landing pages tell you.

[00:05:05]Every Python UI framework is a remote control for something else. Flet drives Flutter. Reflex compiles to React.

[00:05:13]Streamlit drives a React app server-side. NiceGUI drives Vue. PyQt wraps Qt, which is C++. Tkinter wraps Tcl/Tk.

[00:05:24]None of them are Python UIs, they are Python-controlled UIs. Your Python code does not draw anything. It sends instructions to a renderer that structurally was not built for it. In Flet's case, that renderer is Flutter, designed around Dart's reactive widget model. Flet exposes an imperative API on top. You call page.update by hand.

[00:05:45]Developers coming from Flutter call this anti-Flutter philosophy. It's not a bug, it's the abstraction debt of pretending Python can drive Flutter without the impedance showing. Every framework on that list has its own version of the debt. The debt is the product. This is what Flet's existence is actually evidence of. Python has, over 15 years, conquered surface after surface. Data science, machine learning, web backends, infrastructure automation, and in the last 24 months, even its own tooling layer. A company called Astral, now part of OpenAI, built UV and Ruff in Rust.

[00:06:17]Tools that do what pip, virtualenv, black, and flake8 do, but fast enough that developers stopped arguing about the alternatives. Python kept the productivity surface. Rust does the work underneath. Flet is the same trick applied to UI. Python on top. Fast iteration, familiar syntax, the ecosystem you already have. Flutter on the bottom. Actually capable of drawing pixels at 60 frames per second on six different operating systems. The productivity surface stays Python. The rendering layer stays whatever can actually render. This works. 7 million people have downloaded Flet. 16,000 started. The API is stable. Real apps have shipped to real app stores. It also doesn't work cleanly. The bundle size is large because the Flutter runtime ships with every app. Performance is fine for forms and dashboards. Painful for animation or real-time work. Custom components are limited. No rich text editor. No real-time video. And the Pyodide web mode runs Python on a single browser thread. So, any call freezes the UI you just shipped. This is the trade.

[00:07:19]And the trade is not unique to Flet.

[00:07:21]It's the structural shape of every Python UI framework. Each one is a bet on which abstraction debt is least painful for the kind of app you're building. If you tried Flet and gave up because the bundle was huge or the page.update calls felt wrong, you weren't doing it wrong. You hit the abstraction debt. That's the product.

[00:07:39]What does this leave you with? Pick Flet if you need true multi-platform from Python and your app is form-heavy, internal, dashboard-shaped, or MVP stage. That's a real and growing category. It's not anything. It's a specific tier. Pick Reflex if you only need web and want React under the hood.

[00:07:57]Pick Streamlit if your audience is data scientists who will never install a binary. Pick Nice GUI if you're building an admin panel that lives behind your firewall. Pick Flutter native in Dart if you're building a consumer app where bundle size, animation, and 60 FPS scrolling matter more than which language you write in. Flet's existence is evidence of two things at once.

[00:08:18]Python's reach is structural, not accidental. There's enough demand for Python everywhere to fund five competing frameworks. And there's a tier of problems Python will probably never own cleanly. The ones where the rendering, not the logic, is the product. Knowing which tier your project is in is the whole question. Flet is one answer. If you want the analytical version of how Python's own tooling layer got quietly rewritten in Rust, the bigger story about who's eating whose ecosystem, there's a video on the screen now. That one's about Rust. Watch it next. See you there.