1. 1

I’m building an open-source workflow engine on Kubernetes.

The main goal is to make long-running and realtime workflows first-class Kubernetes resources instead of external orchestration state.

I wanted:

  • workflows as CRDs,
  • durable execution,
  • native event triggers,
  • and support for realtime AI/media pipelines (voice, streaming, LiveKit).

Existing tools like Argo worked well for batch DAGs, but realtime streaming workflows felt awkward, especially when combining Kubernetes events and bidirectional media streams.

The project currently supports:

  • cron/webhook/Kubernetes triggers,
  • streaming pipelines,
  • LiveKit voice agents,
  • resumable workflow state,
  • reusable workflow components.

I’d especially appreciate feedback on:

  • whether workflows-as-CRDs is the right abstraction,
  • realtime execution design,
  • and where this overlaps/conflicts with existing workflow systems.

Realtime Demo: https://www.youtube.com/watch?v=b8vu0J1kJ-c

  1.