Hi everyone,
It’s been a while since our last announcement for v0.1.1 in February. The Agent Sandbox subproject has been moving fast, and I’m excited to share a roundup of our recent progress, as well as a new post on the Kubernetes blog!
Justin Santa Barbara and I recently published a blog post: Running Agents on Kubernetes with Agent Sandbox. It covers the architectural shift towards long-running, stateful AI agents and how the agent-sandbox project introduces the right abstractions (like the Sandbox CRD, strong isolation, and SandboxWarmPool) to orchestrate these workloads natively on Kubernetes.
Since v0.1.1, we have released several major versions (currently at v0.4.3) with a strong focus on security, storage, and developer experience. Key highlights include:
SandboxTemplate .volumeClaimTemplates to SandboxTemplate for per-instance persistent storage. We also introduced a Finished condition and a ttlSecondsAfterFinished field for the automatic cleanup of finished claims.client-gen, allowing Go developers to easily interact with Agent Sandbox resources.SandboxWarmPool now creates full Sandbox CRs instead of bare pods, with significantly improved correctness and adoption logic.Getting Started You can find the installation instructions for the latest release, along with our new getting started guides, examples, and expanded documentation, on our GitHub repository and our Docs site.
We'd love to see you get involved: