The backend side of AI agents is getting commoditized fast — LangChain, CrewAI, Vercel AI SDK, dozens of others. The frontend has options too, but they're either React-only (CopilotKit, Vercel AI SDK), full applications you can't decompose (Chatbot UI), or missing AI-specific controls beyond basic chat. If you're not on React, or you need model selectors, parameter panels, generation controls — you're mostly on your own.
I think there's a missing layer: framework-agnostic UI components that connect to any of these backends, so you can go from agent to working product without rebuilding the same chat UI for the hundredth time. Loquix is the first step.
It's 35 Web Components (Lit + TypeScript) covering the full surface — not just messages and input, but generation controls with pause/resume, inline editing, drag-and-drop attachments, model/mode selectors, parameter panels, template pickers, welcome screens.
Web Components specifically because this layer shouldn't dictate your stack. These work in vanilla JS, React, Vue, whatever. Everything communicates through composed custom events, so wiring up any backend is straightforward.
A few things I'm happy with: streaming handles pause/resume properly (not just stop); 100+ CSS variables for theming with zero style leakage; and real accessibility — ARIA, keyboard nav, live regions, axe-core validation. 896 tests, TypeScript strict, MIT licensed.
Next up: backend provider adapters — plug in your LangChain chain or Vercel AI stream and have the UI just work.
The backend side of AI agents is getting commoditized fast — LangChain, CrewAI, Vercel AI SDK, dozens of others. The frontend has options too, but they're either React-only (CopilotKit, Vercel AI SDK), full applications you can't decompose (Chatbot UI), or missing AI-specific controls beyond basic chat. If you're not on React, or you need model selectors, parameter panels, generation controls — you're mostly on your own.
I think there's a missing layer: framework-agnostic UI components that connect to any of these backends, so you can go from agent to working product without rebuilding the same chat UI for the hundredth time. Loquix is the first step. It's 35 Web Components (Lit + TypeScript) covering the full surface — not just messages and input, but generation controls with pause/resume, inline editing, drag-and-drop attachments, model/mode selectors, parameter panels, template pickers, welcome screens.
Web Components specifically because this layer shouldn't dictate your stack. These work in vanilla JS, React, Vue, whatever. Everything communicates through composed custom events, so wiring up any backend is straightforward. A few things I'm happy with: streaming handles pause/resume properly (not just stop); 100+ CSS variables for theming with zero style leakage; and real accessibility — ARIA, keyboard nav, live regions, axe-core validation. 896 tests, TypeScript strict, MIT licensed.
Next up: backend provider adapters — plug in your LangChain chain or Vercel AI stream and have the UI just work.
Repo: https://github.com/loquix-dev/loquix
What's the biggest pain point you're hitting right now when building AI chat UIs?