Google Donates OpenRL to Cloud Native Foundation
The open-source reinforcement learning project was introduced at KubeCon + CloudNativeCon North America and is intended to simplify agent training on Kubernetes.
Updated on Nov. 12, 2026 in Artificial Intelligence

Google has donated its OpenRL project to the Cloud Native Computing Foundation. Engineers Sunil Arora and Justin Santa Barbara introduced the project at KubeCon + CloudNativeCon North America.
Why it matters
Fine-tuning AI models through reinforcement learning is increasingly taxing AI infrastructure, and OpenRL aims to separate that compute plumbing from experimental research. Placing the project under the Cloud Native Computing Foundation opens it to broader community development.
OpenRL abstracts infrastructure into core primitives and enables self-hosted training APIs on Kubernetes clusters. Its Agent Substrate component deploys isolated sandboxes used to calculate rewards during agent training.
The players
Google is a multinational technology company that develops search, cloud computing and artificial intelligence products.
Cloud Native Computing Foundation
The Cloud Native Computing Foundation is an open-source organization that hosts cloud-native software projects, including Kubernetes.
The details
OpenRL includes Agent Substrate, which runs isolated sandboxes for reward calculation during agent training, and the kubectl-ai tool, which uses sandboxing to prevent system damage from autonomous agents. Google's GKE Labs team places experimental projects in users' hands early.
Timeline
Article on the donation was published on November 12, 2026.
OpenRL was introduced at KubeCon + CloudNativeCon North America.
Tech Race
OpenRL extends Kubernetes, the container orchestration system the Cloud Native Computing Foundation stewards, into reinforcement learning training. The move places AI agent training within the same open ecosystem used for general cloud infrastructure.
Developers training AI agents on Kubernetes may gain a self-hosted option for reinforcement learning workflows. Open-source availability through the foundation could lower adoption barriers for teams that run their own infrastructure.
The takeaway
Separating training infrastructure from experimental research lets teams iterate on AI agents without rebuilding their compute stack. Watch for community contributions as the project moves under foundation governance.
Further reading
Explore more on Artificial Intelligence developments in this space.



