The framework AI coding agents loveVS Code ExtensionBeta

Stop splitting backend and frontend

Python server logic and React UI live in one .pyxl file. SSR, file routing, server actions — zero glue code.

pages/index.pyxl
@server
async def load(request):
    return {"user": await db.get_user(request)}

export default function Page({ data }) {
    return <h1>Hello, {data.user.name}</h1>
}

Why Pyxle

One file replaces your entire stack

No more context-switching between Python backend and React frontend. Pyxle unifies them in a single developer experience.

Problem

Backend + frontend

Traditional

2 repos, REST API glue, separate deploys

With Pyxle

One .pyxl file per route

Problem

Data fetching

Traditional

fetch(), loading states, error handling

With Pyxle

@server loader → data becomes React props

Problem

Server mutations

Traditional

API routes + fetch + form handling

With Pyxle

@action + useAction() — one line

Problem

Deployment

Traditional

Build frontend, deploy backend, configure CORS

With Pyxle

pyxle build && pyxle serve

One file. Full stack.

Server + Client in a single .pyxl

Your async Python loader and React component live together. No API wiring, no boilerplate.

pages/dashboard.pyxl
# pages/dashboard.pyxl

from pyxle.runtime import server

@server
async def load_dashboard(request):
    user = await db.get_user(request.state.user_id)
    stats = await db.get_stats(user.id)
    return {"user": user, "stats": stats}


import React from 'react';
import { Head, Link } from 'pyxle/client';

export default function Dashboard({ data }) {
    const { user, stats } = data;
    return (
        <main className="p-8">
            <Head>
                <title>Dashboard</title>
                <meta name="robots" content="noindex" />
            </Head>
            <h1>Welcome back, {user.name}</h1>
            <div className="grid grid-cols-3 gap-4">
                {stats.map(s => (
                    <div key={s.label} className="card">
                        <span>{s.value}</span>
                        <span>{s.label}</span>
                    </div>
                ))}
            </div>
            <Link href="/settings">Settings</Link>
        </main>
    );
}

Features

Everything you need to ship

Pyxle gives you the full stack in one cohesive toolkit. No glue code needed.

File-based routing

Drop .pyxl files into pages/ and get routes instantly. Dynamic segments, catch-all routes, layouts, and route groups.

Async server loaders

Fetch data with @server functions on Starlette. Return a dict and it becomes React props automatically.

Server-side rendering

Every page is server-rendered then hydrated. Fast first paint, great SEO, smooth interactivity.

Vite-powered dev

Instant hot module reloading via Vite. Tailwind CSS built in. Sub-second feedback loop.

Server actions

Mutate data with @action decorators. Call from React via useAction() — type-safe POST handlers, zero glue code.

Production ready

CSRF protection, CORS, middleware, and hashed asset builds. Deploy anywhere Python runs.

Performance

Built for production speed

Full-stack features without the performance tax. Pyxle delivers SSR, file routing, and server actions while outperforming traditional Python frameworks.

1,100+req/s

SSR Throughput

Comparable to Next.js

20k+req/s

POST Handling

1.8x faster than FastAPI

2msp50

Latency

JSON serialization

0errors

Under Load

100 concurrent connections

Python Framework Comparison

Average requests/second across all API tests

Pyxle is the only Python framework with React SSR
FastAPI
8,360 req/s
Pyxle
6,256 req/s
Flask
3,788 req/s
Django
2,787 req/s

Benchmarked with autocannon (10 connections, 12s). Apple M3, Python 3.13, Node.js 24. All frameworks run production configs. See full methodology and results.

Architecture

How Pyxle works

Write one file. Pyxle compiles, serves, renders, and hydrates automatically.

.pyxl File

Python + React in one file

Compiler

Splits into server + client

Starlette

Runs your Python loader

SSR + Hydrate

Full HTML, then interactive

For AI coding agents

The framework AI agents love

Python backend. React frontend. One file per feature. Tiny CLI surface, zero magic, grep-parseable errors. Built so Claude, Cursor, and Copilot can ship real features with a fraction of the context, files, and cross-language glue that traditional full-stack frameworks demand.

Typical stack

4–7 files

per feature. Two languages.

Pyxle

1 file

Python + JSX colocated.

Python, the language LLMs know best

Every frontier LLM has seen more Python than any other language. Your agent stays in its highest-confidence mode for the entire backend — and the entire AI SDK ecosystem (Anthropic, OpenAI, LangChain, instructor) is Python-first anyway.

One file per feature, zero context-switching

Loader, mutation, and React component in a single .pyxl file. An agent reads one file to understand the whole feature — no cross-file consistency risk, no TypeScript/Python type-sync drama, no 7-file edits for a CRUD screen.

Five commands, zero magic

init, dev, check, build, serve. That's the whole vocabulary. Decorators set one attribute and return — no DI, no metaclass, no runtime patching. The code you see is the code that runs, which means the agent's mental model matches reality.

Machine-parseable errors from pyxle check

[section] line N: message, one file per line, exit code 0 or 1. Every diagnostic is grep-friendly. One bad file never aborts the scan, and downstream JSX cascades are suppressed automatically. Your agent's edit → check → fix loop stays tight and high-signal.

File-based routing = agent-friendly config

ls pages/ tells the agent every route in the project. No urls.py to maintain, no router.ts to keep in sync. The filesystem is the configuration — which means zero tokens spent reading routing files.

Fully documented architecture

11 architecture deep-dives covering the parser, compiler, SSR, routing, build, runtime, and CLI. Point your agent at /docs/architecture once and it has the whole mental model — no need to grep source code to answer framework questions.

Use Cases

Your Python. Your React UI.
One framework.

Build AI dashboards, internal tools, and data apps with the Python libraries you already know.

AI Dashboards

Connect your ML models directly. Load predictions in @server, visualize in React. No API layer needed.

@server
async def load(req):
    predictions = model.predict(data)
    return {"results": predictions}

Internal Tools

Admin panels, CRUD apps, data viewers. Build in hours with Python, not days with separate frontend.

@action
async def approve(req):
    body = await req.json()
    await db.update(body["id"])
    return {"ok": True}

Data Apps

Pandas, NumPy, scikit-learn — use your entire Python ecosystem. The UI is just React on top.

@server
async def load(req):
    df = pd.read_csv("data.csv")
    summary = df.describe().to_dict()
    return {"stats": summary}

Get started

Up and running in seconds

Install from git, scaffold a project, and start the dev server.

Install Pyxle

$ pip install pyxle-framework

Create a project

$ pyxle init my-app && cd my-app

Install dependencies

$ pyxle install

Start building

$ pyxle dev
Actively shipping

Stay in the loop

Get notified about new releases, features, and the occasional deep dive into building with Pyxle.

No spam, ever. Unsubscribe anytime.

This form is a live @action demo — powered by Pyxle.