NPM Library: Interactive Chat UI + LLM Compiler for Quick Actions

What it is

A frontend library that upgrades normal AI chat into interactive conversations where the AI can output UI components (buttons, forms, selections, cards) and users can take actions instantly - without re-prompting.

This is a "chat-to-app" interface layer.

What it solves

Text-only chat has major friction:

• Users waste time writing prompts that should be clicks
• Repeated "clarify" loops waste tokens and attention
• Complex flows become messy and inconsistent
• Developers struggle to maintain structured UX in a chat window

This library solves: prompt fatigue + slow UX + lack of structured actions.

How it works

1. AI outputs a structured UI instruction

Instead of only returning text, the AI can output:

• A button group
• A dropdown
• A form
• An action card
• A progress state

2. The library renders it into UI

The library converts that instruction into real UI in the chat.

3. Two-way binding / quick actions

When the user clicks a UI element, the library sends a structured event back:

• "user selected option X"
• "user approved action"
• "form submitted with fields"

4. Developer hooks

Developers can attach handlers to actions by ID:

• AI generates a button with id #dtol
• Developer maps it to "switch to dark theme"

So the AI can propose actions; the app executes them safely.

Key features

Interactive components

• Buttons, quick replies, option chips
• Forms (multi-field inputs)
• Cards (structured content blocks)
• Confirm dialogs for approvals
• Progress + status elements for long workflows

Token-efficient "compiler format"

• Supports compact markup that is cheaper than verbose JSON/HTML
• Frontend compiler converts it into UI elements
• Makes "UI generation" cheaper in tokens and easier to control

Safe action execution

• Allowlist which actions the AI can trigger
• Approval gates for sensitive actions
• Prevents arbitrary script/HTML injection

Portable integration

• Works with any LLM backend (because it's just a rendering + event contract)
• Lets developers keep their own agent stack while upgrading UI

White-label compatibility

• The UI library can match themes for any product
• Works as a core layer inside other platforms

Primary audiences

• Any AI product team building chat experiences
• SaaS products adding assistants with real UI actions
• Developers who want "minimal prompting, maximum actions"

Differentiator

Most chat UIs are message lists.

This is an interaction contract that makes chat behave like a real product UI.