Ross ROSS = Recommend OSS · open-source software intelligence for agents

kubernetes-sigs/agent-sandbox

agent-sandbox enables easy management of isolated, stateful, singleton workloads, ideal for use cases like AI agent runtimes and reinforcement learning (RL). observed · 2026-08-28

github.com/kubernetes-sigs/agent-sandbox · homepage · Go · Apache-2.0 (permissive) observed · 2026-08-28

Health v2 · maintenance only

84/100

  • Activity 99
  • Release rhythm 98
  • Longevity 27
How is this computed?

round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10) — computed 2026-09-03. Adoption (stars, forks) is never an input.

  • gap_med: 11.0
  • age_days: 386
  • days_rel: 13
  • days_push: 7
  • n_releases_24m: 15

Full methodology

Adoption not part of the score

3640 stars · 466 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

Agent Sandbox is a Kubernetes SIG project providing a Sandbox Custom Resource Definition (CRD) and controller for managing isolated, stateful, singleton workloads with stable identity and persistent storage. It is purpose-built for AI agent runtimes, code execution of untrusted/LLM-generated code, and reinforcement learning workloads, with Python and Go SDKs and extension CRDs like SandboxTemplate, SandboxClaim, and SandboxWarmPool.

Use cases

  • run untrusted LLM-generated code in isolated sandboxes on Kubernetes
  • host coding agents that write and execute code with full dev tooling
  • provision pre-warmed sandbox pools for low-latency agent sessions
  • run reinforcement learning and SWE-bench-style evaluation workloads
  • give AI agents isolated browser or desktop environments for computer-use tasks
  • manage stateful singleton pods with stable identity and persistent storage
  • pause and resume sandboxes to save compute costs between sessions

When to choose

  • you need a Kubernetes-native, declarative API for isolated agent execution environments
  • you want millisecond sandbox provisioning via warm pools for high-throughput agent pipelines
  • you need configurable isolation depth with gVisor or Kata Containers runtimes
  • you run RL training or evaluation jobs that require many ephemeral stateful environments

When to avoid

  • you are not running on Kubernetes and cannot operate a cluster
  • you only need simple local code-execution sandboxing without orchestration
  • you need fully replicated stateless workloads, which Deployments already handle well

Facets

framework · maturity active

agent-framework container-orchestration security deployment sdk workflow-automation artificial-intelligence cloud-computing security developer-tools reinforcement-learning go python cloud self-hosted kubernetes-operator crd sandbox code-execution gvisor kata-containers warm-pool untrusted-code computer-use swe-bench ai-agents containers kubernetes docker

10 sources

Member repositories

RepositoryRoleHealth v2
kubernetes-sigs/agent-sandboxmain84

For agents

markdown · JSON · MCP: product_card(name="kubernetes-sigs/agent-sandbox")

Data as of 2026-08-30T08:39:29.467469+00:00 · Report a problem