Frontend & real-time

AI products live or die by how fast they feel. When an agent generates code, the user needs to see tokens streaming in real time. When teammates collaborate on a document, edits need to appear instantly. When a dashboard monitors production systems, data can’t be seconds stale. The frontend infrastructure behind these interactions is a specialty—not something you bolt on at the end.

Zero-latency interfaces

Every spinner, every loading state, every moment your product feels sluggish—that’s a user losing flow state. For professional tools where people spend hours per day, latency isn’t a UX issue. It’s a retention issue.

We build zero-latency interfaces using local-first principles: data lives on the device first, syncs in the background, and works offline by default. Optimistic UI patterns make server-rendered apps feel instant. WebAssembly runs heavy computation in the browser without server round-trips. bolt.new—$40M+ ARR, the fastest-growing AI product ever launched—runs a full Node.js environment inside the browser via WebAssembly. No server round-trips for builds, installs, or file operations.

Dev tool interfaces

Developer tools demand interfaces that handle high information density, complex interactive workflows, and keyboard-first navigation. We build the frontend engineering behind these products.

For Tines, we built a React frontend with interactive node diagrams and real-time state management across hundreds of connected actions, delivering 100x performance improvements in target areas. Tines reached unicorn status. For Tegon, we engineered the React and Next.js frontend for a YC-backed AI-first issue tracker. For HTTPie, we built the desktop app with React, TypeScript, and Electron.

We maintain KeyUX, an open source library that adds keyboard navigation to web apps: hotkey support via aria-keyshortcuts, arrow navigation in menus, and focus management. These are the interaction patterns professional tools need but most teams skip.

Small libraries

The modern frontend ecosystem ships megabytes of JavaScript for routine tasks. We started a small libraries movement to push back: purpose-built packages that solve one problem in as few bytes as possible.

Nano ID generates unique string IDs in 116 bytes—130 times smaller than uuid. It has 111M weekly npm downloads, more than React itself. Nano Stores is a tiny state manager (334 bytes) that works across React, Vue, Svelte, and vanilla JS. size-limit enforces bundle size budgets in CI. If a dependency pushes your bundle past the limit, the build fails before the PR merges. We also maintain PostCSS (158M weekly downloads) and Browserslist (118M weekly downloads)—both bigger than React.

This isn’t just open source advocacy. When we build frontends for clients, we reach for the smallest tool that solves the problem. Fewer bytes mean faster loads, less complexity, and lower infrastructure costs.

Design systems for agentic development

When AI coding tools generate UI, they’re only as good as the component system they work with. A well-structured design system with clear component APIs, typed props, and documented patterns produces consistent agent output. A messy one produces slop.

We build component architectures that work for both humans and agents. In Rails, our view_component-contrib gem standardizes ViewComponent interfaces across the application. In React and TypeScript, we build typed component libraries with explicit prop APIs. The component catalog doubles as a specification that agents follow—the same principle behind AI Harnesses applied to the view layer.

Real-time collaboration

For Akeero, an AWS security audit platform, we built real-time multi-user editing with Logux—our open source library for local-first sync. Edits appear instantly for all participants, with conflict resolution handled automatically.

For Healthie, a virtual-first healthcare SaaS, we migrated their real-time features to AnyCable for better performance and reliability. For ClickFunnels, we stabilized their AnyCable infrastructure at scale. We’ve solved WebSocket scaling, connection avalanches, and delivery guarantees at production scale.

LLM streaming and generative UI

Streaming AI responses over HTTP works until it doesn’t. Token-by-token delivery over WebSockets solves the problems HTTP streaming creates: dropped connections, backpressure handling, multiplexing multiple streams, and delivering structured UI updates alongside raw tokens. We built this infrastructure for AI products and open sourced the patterns through AnyCable.

Voice AI

Voice AI needs real-time infrastructure with stricter latency requirements than chat or collaboration—bidirectional audio streaming, speech-to-text processing, and sub-second response times. We’ve built this stack with AnyCable.

Doximity, used by 80% of US doctors, ships voice features powered by AnyCable—including “Hold for Me,” where a digital assistant handles phone holds and alerts the doctor when a real person picks up. We built a voice assistant connecting Twilio with OpenAI’s Realtime API through AnyCable Go—bidirectional audio, speech-to-text, and tool calling over WebSockets. We’ve also connected Twilio streams with offline speech recognition for real-time call transcription. LiveVoice, an ultra-low-latency audio streaming platform for simultaneous interpretation, runs on AnyCable Pro.

When you need this

Zero-latency interfaces, devtool dashboards, interactive workflow builders, real-time collaboration, LLM streaming, voice AI, design systems, component architectures. If your product needs a frontend that’s fast, interactive, and built for daily professional use, we’ve done it before.