Microsoft Research Asia releases small toolkit for training production agents with reinforcement learning
Why it matters
You can train agents inside the harness you already deploy, on your own Kubernetes, without rebuilding them or buying a sandbox service.
Ask your team
Could we train our deployed agents on our own Kubernetes cluster instead of rebuilding them in a separate training framework?
What happened
Microsoft Research Asia released Agent Lightning v1.0, a roughly 3,500-line framework that trains agents with using the same harness they run in production. It runs agents as standard Kubernetes jobs on your own cluster, cloud Kubernetes or local machines, with no paid sandbox service.
Tap an underlined word to see what it means.
Impact on AI & developer tools · Software & SaaS
The context Last Microsoft story: Microsoft opens first Southeast Asia research lab in Singapore on July 24 2025 (3 Oct)
Read the original at Microsoft Research blog
Something wrong in this story? Report a mistake. We check it against the source within 24 hours.
Share with your take
Start from our line and make it yours. It stays on your device until you post it.
Ask about this story
Answered only from this story, with the line it came from.