Built by engineers who kept rebuilding the same thing.

The problem we kept running into

Every time we shipped a new agent feature — whether a customer-facing chat, an internal analyst tool, or a workflow automation — we found ourselves rebuilding the same UI primitives.

A message thread. A trace viewer so users could see what the model actually did. An approval card for actions that needed human sign-off. A feedback widget for thumbs up/down. Session history so users could pick up where they left off.

None of it was the hard part. All of it took weeks. And it always looked like it was built in a hurry, because it was.

The design system that didn't exist

shadcn/ui gives you buttons and form inputs. Tailwind gives you utilities. Neither gives you a ToolCallTrace component that handles streaming, expandable steps, timing, and error states out of the box.

Agent Interface is the component library that should have existed. We built it for ourselves first. Now we're opening it to everyone.

What we believe

The interface between a human and an AI agent is its own design problem. It's not a chat app. It's not a dashboard. It's something new — and it deserves components designed specifically for it.

Transparency matters. Users should be able to see what an agent did, not just what it said. Approval flows should feel safe, not alarming. Feedback should be frictionless enough to actually collect.

Open source is the right model for component libraries. The core components are MIT-licensed and will stay that way. We make money on hosted infrastructure, not on the primitives that every team needs.