Creating Custom UI Components

marimo provides a rich library of built-in UI components, but you can also create custom components using anywidget - a framework for building custom Jupyter widgets with modern web technologies.

Why Custom Components?

Create custom components when you need:

Specialized visualizations not covered by built-in components
Integration with third-party JavaScript libraries
Custom interactive controls for domain-specific workflows
Shareable, reusable components across projects

Using anywidget with marimo

marimo provides seamless integration with anywidget through mo.ui.anywidget():

import marimo as mo
from my_widget import MyCustomWidget

# Wrap anywidget with mo.ui.anywidget()
widget = mo.ui.anywidget(MyCustomWidget())

# Access widget state
widget.value  # Returns widget's current state as a dictionary

Create a simple color picker widget:

import anywidget
import traitlets

class ColorPicker(anywidget.AnyWidget):
    _esm = """
    export function render({ model, el }) {
        let input = document.createElement('input');
        input.type = 'color';
        input.value = model.get('color');
        
        input.addEventListener('input', () => {
            model.set('color', input.value);
            model.save_changes();
        });
        
        el.appendChild(input);
    }
    """
    
    color = traitlets.Unicode('#3b82f6').tag(sync=True)

# Use in marimo
import marimo as mo
color_picker = mo.ui.anywidget(ColorPicker())

# In another cell, access the selected color
mo.md(f"Selected color: {color_picker.color}")

Component Architecture

Anatomy of an anywidget

From marimo/_plugins/ui/_impl/from_anywidget.py:

class anywidget(UIElement[ModelIdRef, AnyWidgetState]):
    """
    Create a UIElement from an AnyWidget.
    
    This proxies all the widget's attributes and methods, allowing seamless
    integration of AnyWidget instances with marimo's UI system.
    """
    
    def __init__(self, widget: AnyWidget):
        self.widget = widget
        # Widget state is synchronized automatically
        # with marimo's reactive system

Key components:

Python traits: Define the widget’s state (synchronized between Python and JavaScript)
JavaScript module: Renders the UI and handles interactions
marimo wrapper: Integrates with marimo’s reactive execution

State Synchronization

marimo automatically synchronizes widget state:

# Changes in JavaScript automatically update Python
widget.color = '#ff0000'  # Updates both Python and frontend

# Access current state
current_state = widget.value  # Dict of all traits

From examples/third_party/anywidget/tldraw_colorpicker.py:

import marimo as mo
import numpy as np
import matplotlib.pyplot as plt
from tldraw import ReactiveColorPicker

# Create the widget
color_widget = mo.ui.anywidget(ReactiveColorPicker())

# Use in visualization
fig, ax = plt.subplots()
ax.scatter(x, y, s=sizes*5, color=color_widget.color or None)

mo.hstack([color_widget, plt.gca()])

import anywidget
import traitlets

class RangeSelector(anywidget.AnyWidget):
    _esm = """
    export function render({ model, el }) {
        const container = document.createElement('div');
        
        const minInput = document.createElement('input');
        minInput.type = 'range';
        minInput.min = 0;
        minInput.max = 100;
        minInput.value = model.get('min_value');
        
        const maxInput = document.createElement('input');
        maxInput.type = 'range';
        maxInput.min = 0;
        maxInput.max = 100;
        maxInput.value = model.get('max_value');
        
        minInput.addEventListener('input', () => {
            model.set('min_value', parseInt(minInput.value));
            model.save_changes();
        });
        
        maxInput.addEventListener('input', () => {
            model.set('max_value', parseInt(maxInput.value));
            model.save_changes();
        });
        
        container.appendChild(minInput);
        container.appendChild(maxInput);
        el.appendChild(container);
    }
    """
    
    min_value = traitlets.Int(0).tag(sync=True)
    max_value = traitlets.Int(100).tag(sync=True)

# Use in marimo
range_selector = mo.ui.anywidget(RangeSelector())

# Access both values
filtered_data = data[
    (data > range_selector.min_value) & 
    (data < range_selector.max_value)
]

Advanced Features

Using External Libraries

Integrate JavaScript libraries like D3, Plotly, or custom frameworks:

import anywidget
import traitlets

class D3Visualization(anywidget.AnyWidget):
    _esm = """
    import * as d3 from "https://cdn.jsdelivr.net/npm/d3@7/+esm";
    
    export function render({ model, el }) {
        const data = model.get('data');
        
        const svg = d3.select(el)
            .append('svg')
            .attr('width', 400)
            .attr('height', 300);
        
        svg.selectAll('circle')
            .data(data)
            .enter()
            .append('circle')
            .attr('cx', d => d.x)
            .attr('cy', d => d.y)
            .attr('r', 5)
            .on('click', (event, d) => {
                model.set('selected_point', d);
                model.save_changes();
            });
    }
    """
    
    data = traitlets.List([]).tag(sync=True)
    selected_point = traitlets.Dict({}).tag(sync=True)

CSS Styling

Add custom styles to your widget:

class StyledWidget(anywidget.AnyWidget):
    _esm = """
    export function render({ model, el }) {
        const button = document.createElement('button');
        button.textContent = model.get('label');
        button.className = 'custom-button';
        el.appendChild(button);
    }
    """
    
    _css = """
    .custom-button {
        background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
        color: white;
        border: none;
        padding: 12px 24px;
        border-radius: 8px;
        cursor: pointer;
        font-size: 16px;
        transition: transform 0.2s;
    }
    
    .custom-button:hover {
        transform: scale(1.05);
    }
    """
    
    label = traitlets.Unicode("Click me").tag(sync=True)

Binary Data Transfer

Transfer binary data efficiently (images, arrays, etc.):

import anywidget
import traitlets
import numpy as np

class ImageWidget(anywidget.AnyWidget):
    _esm = """
    export function render({ model, el }) {
        const canvas = document.createElement('canvas');
        const ctx = canvas.getContext('2d');
        
        model.on('change:image_data', () => {
            const data = new Uint8ClampedArray(model.get('image_data'));
            const imageData = new ImageData(data, model.get('width'));
            canvas.width = model.get('width');
            canvas.height = model.get('height');
            ctx.putImageData(imageData, 0, 0);
        });
        
        el.appendChild(canvas);
    }
    """
    
    image_data = traitlets.Bytes().tag(sync=True)
    width = traitlets.Int(256).tag(sync=True)
    height = traitlets.Int(256).tag(sync=True)

# Use with numpy array
img_widget = mo.ui.anywidget(ImageWidget())
img_array = np.random.randint(0, 255, (256, 256, 4), dtype=np.uint8)
img_widget.image_data = img_array.tobytes()
img_widget.width = 256
img_widget.height = 256

From marimo/_plugins/ui/_impl/anywidget/init.py:

class CommLifecycleItem(CellLifecycleItem):
    """Manages widget cleanup when cells are re-executed."""
    
    def dispose(self, context: RuntimeContext, deletion: bool) -> bool:
        # Close the communication channel
        self._comm.close()
        return True

marimo automatically:

Initializes widgets when cells run
Cleans up widgets when cells are re-executed or deleted
Manages communication channels between Python and JavaScript

Best Practices

Keep state minimal

Only sync essential state between Python and JavaScript:

# ✓ Good: Only sync what's needed
class GoodWidget(anywidget.AnyWidget):
    selected_id = traitlets.Unicode().tag(sync=True)

# ❌ Avoid: Syncing large computed values
class BadWidget(anywidget.AnyWidget):
    full_dataset = traitlets.List().tag(sync=True)  # Too much data

Handle initialization properly

Ensure your widget handles initial state correctly:

export function render({ model, el }) {
    // Read initial state
    const initialValue = model.get('value');
    
    // Set up UI with initial state
    const input = createInput(initialValue);
    
    // Listen for changes from Python
    model.on('change:value', () => {
        input.value = model.get('value');
    });
    
    // Send changes to Python
    input.addEventListener('input', () => {
        model.set('value', input.value);
        model.save_changes();
    });
}

Use type hints

Document widget traits with proper types:

from typing import List, Dict

class TypedWidget(anywidget.AnyWidget):
    items: List[str] = traitlets.List(trait=traitlets.Unicode()).tag(sync=True)
    config: Dict[str, any] = traitlets.Dict().tag(sync=True)

Test your widgets

Write tests for widget behavior:

def test_widget_initialization():
    widget = mo.ui.anywidget(MyWidget())
    assert widget.value == expected_initial_value

def test_widget_update():
    widget = mo.ui.anywidget(MyWidget())
    widget.property = "new_value"
    assert widget.value['property'] == "new_value"

Publishing Widgets

Package Structure

my-widget/
├── pyproject.toml
├── README.md
├── src/
│   └── my_widget/
│       ├── __init__.py
│       └── widget.py
└── tests/
    └── test_widget.py

pyproject.toml

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[project]
name = "my-widget"
version = "0.1.0"
description = "Custom marimo widget for ..."
requires-python = ">=3.9"
dependencies = [
    "anywidget",
    "marimo",
]

[project.optional-dependencies]
dev = [
    "pytest",
    "ruff",
]

Publishing to PyPI

# Build the package
python -m build

# Upload to PyPI
python -m twine upload dist/*

Usage After Publishing

# Users install your widget
# pip install my-widget

import marimo as mo
from my_widget import MyWidget

widget = mo.ui.anywidget(MyWidget())

Examples Gallery

import anywidget
import traitlets
import pandas as pd

class DataExplorer(anywidget.AnyWidget):
    _esm = """/* Interactive data exploration UI */"""
    data = traitlets.List().tag(sync=True)
    selected_rows = traitlets.List().tag(sync=True)

# Use with pandas
explorer = mo.ui.anywidget(DataExplorer())
explorer.data = df.to_dict('records')

# Get selected data
selected_df = df.iloc[explorer.selected_rows]

class FormBuilder(anywidget.AnyWidget):
    _esm = """/* Dynamic form creation */"""
    schema = traitlets.Dict().tag(sync=True)
    form_data = traitlets.Dict().tag(sync=True)

form = mo.ui.anywidget(FormBuilder())
form.schema = {
    'name': {'type': 'text', 'label': 'Name'},
    'age': {'type': 'number', 'label': 'Age'},
}

# Access submitted form data
submitted_data = form.form_data

Debugging Widgets

Console Logging

export function render({ model, el }) {
    console.log('Initial state:', model.get('value'));
    
    model.on('change', () => {
        console.log('State changed:', model.attributes);
    });
}

Error Handling

class RobustWidget(anywidget.AnyWidget):
    _esm = """
    export function render({ model, el }) {
        try {
            // Your widget code
        } catch (error) {
            console.error('Widget error:', error);
            el.innerHTML = `<div style="color: red;">Error: ${error.message}</div>`;
        }
    }
    """

anywidget Documentation - Complete anywidget reference
marimo UI Elements - Built-in marimo components
anywidget Examples - Example widgets
Jupyter Widgets - IPython widget ecosystem

Get Started

Core Features

Working with Notebooks

Deployment

Data & Visualization

Editor Features

Advanced Topics

Creating Custom UI Components

Creating Custom UI Components

Why Custom Components?

Using anywidget with marimo

Quick Start: Basic Widget

Component Architecture

Anatomy of an anywidget

State Synchronization

Building a Complete Widget

Example: Interactive Drawing Widget

Widget with Multiple Traits

Advanced Features

Using External Libraries

CSS Styling

Binary Data Transfer

Widget Lifecycle

Best Practices

Publishing Widgets

Package Structure

pyproject.toml

Publishing to PyPI

Usage After Publishing

Examples Gallery

Data Visualization Widget

Form Builder Widget

Debugging Widgets

Console Logging

Error Handling

Build docs developers (and LLMs) love

Get Started

Core Features

Working with Notebooks

Deployment

Data & Visualization

Editor Features

Advanced Topics

​Creating Custom UI Components

​Why Custom Components?

​Using anywidget with marimo

​Quick Start: Basic Widget

​Component Architecture

​Anatomy of an anywidget

​State Synchronization

​Building a Complete Widget

​Example: Interactive Drawing Widget

​Widget with Multiple Traits

​Advanced Features

​Using External Libraries

​CSS Styling

​Binary Data Transfer

​Widget Lifecycle

​Best Practices

​Publishing Widgets

​Package Structure

​pyproject.toml

​Publishing to PyPI

​Usage After Publishing

​Examples Gallery

​Data Visualization Widget

​Form Builder Widget

​Debugging Widgets

​Console Logging

​Error Handling

​Related Resources

Build docs developers (and LLMs) love

Creating Custom UI Components

Why Custom Components?

Using anywidget with marimo

Quick Start: Basic Widget

Component Architecture

Anatomy of an anywidget

State Synchronization

Building a Complete Widget

Example: Interactive Drawing Widget

Widget with Multiple Traits

Advanced Features

Using External Libraries

CSS Styling

Binary Data Transfer

Widget Lifecycle

Best Practices

Publishing Widgets

Package Structure

pyproject.toml

Publishing to PyPI

Usage After Publishing

Examples Gallery

Data Visualization Widget

Form Builder Widget

Debugging Widgets

Console Logging

Error Handling

Related Resources