Production-minded AI/LLM infrastructure on Kubernetes.
Open engineering guides, deployment tooling, and integrations for teams running training, inference, RAG, and AI agents on Kubernetes.
| Area | Focus |
|---|---|
| AI/LLM on Kubernetes | Architecture and field guidance for accelerators, scheduling, networking, storage, training, inference, RAG, and platform operations. |
| Agent runtime and sandboxing | Policy-controlled execution, workspace isolation, browser automation, and secure tool integration. |
| Deployment and packaging | Reproducible containers, Helm charts, upgrade paths, release tracking, and multi-architecture artifacts. |
| Production operations | Benchmarks, observability, reliability, security, capacity planning, and incident-driven practices. |
| Project | What it does | Stack |
|---|---|---|
cubesandbox-agent-adapter |
Policy-controlled CubeSandbox execution adapter for OpenClaw and DeepSeek Harness. | Python · microVM · AI agents |
- Build an AI platform: explore the AIK8s infrastructure knowledge base.
- Integrate a secure agent sandbox: start with
cubesandbox-agent-adapter. - Contribute: read the contribution guide and work in the repository closest to your change.
- Get support: follow the support guide; report security issues privately as described in the security policy.
- Production before demos: define operating boundaries, failure modes, and verification steps.
- Reproducible by default: pin versions, publish immutable artifacts, and document upgrades and rollback.
- Secure execution: use least privilege, explicit trust boundaries, and isolated workloads.
- Evidence over opinion: tie recommendations to upstream documentation, real tests, or measurable trade-offs.
中文简介
AIK8s 聚焦运行在 Kubernetes 上的 AI/LLM 基础设施，持续整理 GPU 与异构算力、调度、网络、存储、训练、推理、RAG、Agent Sandbox、可观测性、安全与生产运维实践。
这里不仅沉淀架构知识，也维护可复现的容器制品、部署工具和开源集成。我们优先说明版本边界、验证方法、故障路径与上线条件，让内容和项目能够真正用于工程落地。