03
Agentic Runtime · 2026
Lumen
Durable workflow runtime for AI agents that don't break in production.
Live ProjectIn development

Overview
A runtime platform where developers define AI agent workflows as TypeScript functions — Lumen persists every step, survives crashes, and deploys globally with a single CLI command. Built-in observability with per-step traces, replays, and cost attribution.
My role
Full-stack — runtime design, landing page, chat UI, deployment
Highlights
- 01
Durable execution — workflows survive crashes and resume from any step automatically
- 02
Per-step observability with traces, replays, and token cost attribution
- 03
One-command deploy (`lumen deploy`) — no Docker, no Kubernetes
Key features
- Three primitives: durable workflows, inspector, and global deployment target
- First-class AI agent loops with built-in tools and memory support
- Interactive chat UI for workflow interaction
- BYOC (Bring Your Own Cloud) for enterprise — run in your own AWS/GCP/Azure
- Three-tier pricing with a generous free tier (10k step executions/mo)