A Streaming-First UI Framework Built for the AI Era

#webdev #javascript #ai #opensource

Most AI products still stream words.

But what if an AI model could stream the interface itself?

I recently explored Aktion, an ambitious open-source project created by my mentor, Asfandiyar Khan, a senior frontend engineer based in Germany.

Aktion is a streaming-first UI framework designed to turn compact, JavaScript-shaped instructions into reactive and interactive user interfaces.

Instead of producing another Markdown response or a large block of frontend code, an AI model can progressively generate:

Cards
Charts
Forms
Tables
Dashboards
Navigation
Interactive workflows

What makes Aktion different?

Traditional AI-generated frontend code often requires the model to coordinate several independent systems:

Framework syntax
Package imports
State management
Routing
Form handling
Styling
Data fetching
Component libraries
Dependency versions

Aktion takes a different approach.

It gives developers and AI agents:

A compact JavaScript-shaped reactive language
A unified UI runtime
A consistent component vocabulary
Built-in application capabilities
A framework-agnostic Web Component
Streaming support for progressively generated interfaces

The central integration point is the <aktion-app> Web Component.

Because of this approach, Aktion can be embedded inside React, Vue, Angular, Svelte, plain HTML, or another host environment without rebuilding the generated UI layer for every framework.

Why streaming-first UI matters

Most AI interfaces wait until a response is complete before presenting the final output.

Aktion is designed so completed statements can be rendered while the model is still generating the rest of the response.

A heading can appear first.

Then a chart.

Then filters, forms, cards, and interactive controls.

The AI response does not only describe an application.

It gradually becomes the application.

A smaller target for AI-generated interfaces

One of Aktion’s most interesting ideas is its constrained language.

AI models normally have enormous freedom when generating frontend applications. They must choose libraries, understand different APIs, manage imports, and connect tools that were not necessarily designed to work together.

More freedom can also create more incompatible decisions.

Aktion reduces that decision space by providing one language, one runtime, and one predictable set of interface primitives.

The important idea is not only that Aktion contains many features.

It is that AI-generated interfaces may become more dependable when models generate into a smaller and more controlled environment.

More than a component library

Aktion brings several common frontend concerns into one system, including:

Reactive state
Routing
Forms
Data fetching
Themes
Responsive styling
Charts
Realtime interfaces
Internationalization
Testing utilities
Server-side rendering
A large built-in component vocabulary

This does not automatically make it the right solution for every application.

However, it makes Aktion an interesting experiment in how frontend runtimes could evolve for generative UI.

The engineering effort behind Aktion

Building a framework is very different from building a small library.

Aktion brings together:

A language
A parser
A reactive runtime
A rendering system
A component vocabulary
Streaming behavior
Developer tooling
LLM integration
Documentation
Interactive demonstrations

It is inspiring to see my mentor transform years of frontend experience into such an ambitious open-source project.

An idea worth exploring

Aktion may still be an early project, and questions around security, runtime size, extensibility, debugging, and long-term stability deserve careful discussion.

But the architectural question behind it is important:

Should AI continue generating unrestricted frontend applications, or should we design smaller runtimes that give models a clearer and more predictable target?

The future of generative UI may not depend only on making AI models more capable.

It may also depend on designing better environments for them to generate into.