Our take
Activeregular commits and releases- Activity
- 81
- Community
- 94
- Issues
- 72
- Pull requests
- 62
+344 stars in the last 30 days
- Merges more pull requests than 88% of the projects we track
- Releases more often than 78% of the projects we track
- More contributors than 78% of the projects we track
This is an actively maintained Kubernetes-native option for stateful singleton workloads, with recent releases and contributions from a broad contributor base.
Good fit if
- Your workloads run on Kubernetes and need a stable identity plus persistent storage.
- You want declarative lifecycle management for agent, RL, development, or research sandboxes.
- You can select and operate an appropriate runtime for the isolation level you need.
Look elsewhere if
- You need a standalone sandbox manager rather than a Kubernetes controller.
- You expect Agent Sandbox itself to provide low-level container isolation; it delegates that to runtime implementations.
- Your workload fits the standard Deployment or StatefulSet model and does not need the Sandbox lifecycle or extensions.
All health signals
| Last commit | 2026-10-20 (today) |
|---|---|
| Commits, last 90 days | 100+ |
| Releases, last 12 months | 27 (latest v1.0.7, 2026-10-15) |
| Contributors | 100+ (top contributor: 9% of commits) |
| Issues closed, last 90 days | 9+ (typically closed in 1 day) |
| Pull requests merged, last 90 days | 83 (typically merged in 1 day) |
| Project age | 14 months |
Checked on 2026-10-21 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
4,610Forks
615Change since Nov 14, 2026: +6 stars·+1 forks
Overview
Agent Sandbox is built on Kubernetes, using a Sandbox custom resource definition and controller to manage isolated, stateful, singleton workloads as Pods. It addresses workloads that need a stable identity and persistent storage but do not fit neatly into the replicated model of Deployments or the numbered replicas of StatefulSets.
The project is an orchestrator, not a container-isolation runtime: it configures Pods to use Kubernetes RuntimeClass options such as gVisor or Kata Containers. Optional extensions add reusable templates, claims, and pools of pre-warmed Sandboxes.
Key Features
- Manages a single stateful Pod with a stable hostname and network identity.
- Supports persistent storage that can survive restarts.
- Handles Sandbox lifecycle operations, including scheduled deletion, pausing, and resuming.
- Provides
SandboxTemplateresources for reusable configurations. - Offers
SandboxClaimandSandboxWarmPoolextensions for allocating pre-warmed Sandboxes. - Includes Go and Python SDKs, plus an optional HTTP Sandbox Router.
- Integrates with Kubernetes runtime choices through
RuntimeClass; isolation depends on the selected runtime.
Use Cases
- AI platform teams can provision persistent, isolated runtimes for agents executing generated code.
- Reinforcement learning and evaluation teams can use Sandbox claims and warm pools when they need rapid access to execution environments.
- Developer-tool teams can provide isolated, network-accessible workspaces that retain state across restarts.
- Research teams can run persistent single-container sessions, such as notebook environments.
- Operators can manage single-instance services that need a stable identity without configuring a StatefulSet, Service, and storage separately.
What you need
Detected in the repository
- Go 1.26.0 or newer (from go.mod)
- A Dockerfile, so it can run in a container
- A Helm chart for Kubernetes
- A test suite and automated checks on GitHub Actions
License in plain words
Apache-2.0permissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the license, the NOTICE file and a note of your changes
- Share your changes: no
- Includes an explicit patent grant from the contributors:
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
Install the core controller and extensions from the latest release:
kubectl apply -f https://github.com/kubernetes-sigs/agent-sandbox/releases/latest/download/sandbox-with-extensions.yaml
Then create a Sandbox resource in your cluster. See the Getting Started guide and README for setup and examples.
Alternatives
- agent-sandbox: It exposes isolated agent environments through a REST API and MCP server with E2B compatibility, while this project centers on Kubernetes CRDs and a controller.
| Project | Language | License | Stars | Status |
|---|---|---|---|---|
| agent-sandbox | Go | Apache-2.0 | 4.3k | Active |
| Agent-Sandbox | Go | Apache-2.0 | 222 | Active |
Considerations
- This is Kubernetes infrastructure: adopting it requires a cluster and familiarity with custom resources and controllers. The repository also lists Go, Docker, and Helm among its requirements.
- Agent Sandbox coordinates Pods but does not itself supply the isolation boundary. Teams must choose, configure, and assess a runtime such as gVisor or Kata Containers for their threat model.
- The project is relatively young, so teams should validate its APIs and lifecycle behavior against their operational needs. The project is active, with frequent commits and releases, more than 100 contributors, and issues and pull requests being handled regularly.
- Hibernation, automatic resume on network connection, and some storage and identity capabilities are described as desired characteristics or areas of exploration, not guaranteed features.
Found this useful?
Share it with someone who would like agent-sandbox.