We're building the runtime for autonomous software

In 2024 we were running a small AI product and writing duct-tape Python every day to coordinate the agents that ran it. Cron files, Notion docs, Slack channels named "agent-status-please-read." It worked until it didn't.

We wanted something that treated agents as first-class engineering primitives — like how Kubernetes treats containers, or how Vercel treats deploys. Something boring, reliable, and open. We couldn't find it, so we built it.

RunLoop is what we wished we'd had. We open-sourced it because the next generation of AI products will be built by teams much smaller than ours, and they shouldn't have to write the same duct tape we did.

What drives us is simple: we want to be the infrastructure team you'd trust if your entire product depended on it. Because for our users, it does.

We believe great agent products are useless without reliability, and great reliability is impossible without transparency. RunLoop was built to do both. We're not interested in handing you a black-box runtime that breaks in production. We're here to ship code you can read, audit, and own. That's our work. That's our standard.