Rather than calling tools one by one, Orion 2 generates a program and runs it end to end, meaning fewer round-trips and lower latency. You can try it out at https://chat.vlm.run

When orchestration is code, every workflow is composable, inspectable, and deterministic.

We put together a short Orion 2 demo video: https://www.youtube.com/watch?v=rzhXcNAYQ-0