Sympozium: A Kubernetes-Native Agentic Control Plane (CNCF Technical Community Group Meeting)
Sympozium: A Kubernetes-Native Agentic Control Plane
Part of the CNCF Artificial Intelligence Technical Community Group Meeting Series.
In this session of our bi-weekly TCG meeting series, we have the pleasure to host Alex Jones (Principal Engineer, AWS), creator of the CNCF project K8sGPT, to introduce Sympozium — an open-source, Kubernetes-native control plane for running fleets of AI agents in production.
Most agent frameworks dump every tool into a single shared process, leaving operators one stray kubectl delete away from disaster. Sympozium takes a different path: every agent is an ephemeral Pod, every policy is a CRD enforced by an admission webhook, and every execution is a Job with its own ephemeral, least-privilege RBAC that is garbage-collected the moment the run finishes. The result is a multi-tenant, horizontally scalable, safe-by-design substrate for agentic workloads.
In this session, Alex will walk through the architecture, deploy the control plane live, and show how SkillPacks and Ensembles let you grant agents capabilities — cluster diagnostics, code review, or domain-specific tools — without standing god-roles or leftover permissions.
Key Takeaways:
- Kubernetes-Native Agents: How Pods, CRDs, Jobs, and RBAC map onto agentic concepts — identity, skills, policies, and runs.
- Safety by Design: Defence-in-depth through sidecar isolation, network policies, and ephemeral per-run RBAC with no standing privileges.
- Multi-Agent Orchestration: Ensembles of collaborating agents with delegation, sequential pipelines, and shared workflow memory — governed by a synthetic membrane that controls what flows between agents with selective permeability, token budgets, and circuit breakers.
- From K8sGPT to Sympozium: How lessons from a graduated CNCF AI project evolved into a full agentic control plane.
- Live Demo: Installing Sympozium via Helm, onboarding your first Ensemble, and watching agents delegate tasks to each other in real time.
Project Links:
- Website: https://sympozium.ai
- GitHub: https://github.com/sympozium-ai/sympozium
Joining Details 🔑
Joining details for the meeting can be found below; please use your LF username and password to avoid challenges with the joining URL.
Alternatively, use the Zoom login credentials to join the session directly.
Ways To Join The Meeting 🔗
Join from PC, Mac, iPad, or Android.
Meeting ID: 96058530101.
Meeting Passcode: 290975.
International numbers: https://zoom.us/u/alwnPIaVT.