Skip to main content
||6 min read

Cosmonic at KubeCon + CloudNativeCon NA 2026

Cosmonic at KubeCon + CloudNativeCon NA 2026

We're heading to Salt Lake City for KubeCon + CloudNativeCon NA 2026, and we'd love to see you there!

From November 9-12, you can find us at Booth 224 showing sandboxed agents and tools built with the open source Cosmonic Agent SDK on Cosmonic Desktop and Cosmonic Control, running a hands-on workshop on sandboxing agent tools on Kubernetes, and talking portable GPU inference at Cloud Native AI + Inference Day.

Where to find us​

Cosmonic booth​

Tuesday through Thursday, you can find us in the Solutions Showcase (Salt Palace, Level 1) at Booth 224, in Hall 2:

Solutions Showcase map with Cosmonic Booth 224 circled in Hall 2

Solutions Showcase hours (Mountain Time):

  • Tuesday, November 10: 10:45 AM - 5:15 PM, plus KubeCrawl + CloudNativeFest from 6:00 PM - 7:30 PM
  • Wednesday, November 11: 10:30 AM - 5:00 PM
  • Thursday, November 12: 10:30 AM - 2:00 PM

What you'll see at the booth​

AI agents are useful because they act: they run code, call tools, and reach the network. That's also what makes them risky. At Booth 224, we're showing how to give agents and the tools they use a sandbox, so each one gets only the capabilities you grant it, and nothing else.

  • The Cosmonic Agent SDK. Our new open source SDK gives every agent and every tool its own sandbox, small enough to run by the million. Each part of an agent (the model, its tools and its session) has its own typed interface, so you use only the parts you need, and agents work with the OpenAI-compatible models and MCP servers you already use.
  • Sandboxed agents and tools on your laptop. Cosmonic Desktop runs agents, tools, and the code they generate as sandboxed workloads on your own machine. Each workload declares what it can reach (a host, a directory, a secret), and everything else is denied by default.
  • The same sandbox on Kubernetes. Cosmonic Control brings those workloads to your clusters, so the agent sandbox you build locally is the one you run in production.
  • Ultra-dense, scale-to-zero components. Sandboxed components are small and start fast enough to scale to zero when idle. In wasmCloud's published density benchmark, one 64-core server kept 1.42 million sandboxed components warm and ready, and held 9.69 million scaled to zero. Stop by to see it, and ask about our multi-device scaling demo.
  • A GitHub pull request triage agent. Watch an agent do a real job, triaging pull requests on GitHub, with read access it can use and write access it doesn't have.

Come see a live demo, ask us hard questions about agent security, and grab some stickers.

Workshop: Sandboxing on Kubernetes with WebAssembly​

Sandboxing on Kubernetes with WebAssembly: A Hands-on Tutorial - Bailey Hayes & Eric Gregory, Cosmonic
Wednesday, November 11, 4:10 PM - 5:30 PM MST, Hyatt Regency | Level 4 | Regency Ballroom A

In this hands-on workshop, you'll deploy a sandboxed agent harness (the program that drives a model and calls its tools) to your own Kubernetes cluster with wasmCloud, the CNCF project at the heart of Cosmonic Control. Then you'll give the agent familiar tools like curl and sqlite3, try each tool's most famous escapes, and watch them fail. Finally, you'll put the agent to work triaging pull requests on GitHub.

The workshop is designed for engineers and platform operators with basic Kubernetes familiarity. No prior experience with Rust or WebAssembly is required.

Get set up before the session. Conference Wi-Fi can be slow, so please download the toolchains ahead of time. You'll need Docker (Docker Desktop, OrbStack, or Podman) with 4 GB or more of memory on macOS or Linux, or WSL2 on Windows. Clone the workshop repo and run:

git clone https://github.com/cosmonic-labs/sandbox-workshop-kcna26
cd sandbox-workshop-kcna26
make tools doctor up

make tools downloads pinned, checksum-verified tools into the repo (nothing is installed system-wide, and nothing needs sudo), make doctor checks your machine, and make up starts a local kind cluster with wasmCloud. Prefer not to install anything? The workshop also runs in GitHub Codespaces.

Talk: GPU portability with WebGPU​

Cloud Native AI + Inference Day on Monday, November 9​

GPU Portability Was Broken but Now There's WebGPU - Bailey Hayes & Mendy Berger, Cosmonic
1:30 PM - 1:55 PM MST, Salt Palace | Level 2 | 254

Running AI inference on Kubernetes usually means choosing between vendor-specific runtimes, driver lock-in, and operator configurations that break across clouds. Bailey and Mendy will show a different approach: wasi:webgpu, an open, vendor-neutral interface that brings the WebGPU standard to the server. A sandboxed component simply declares GPU access as a capability, and wasmCloud makes it schedulable on Kubernetes.

You'll see ONNX models running inference through wasi:webgpu with no framework-specific operator or driver assumptions, so the same component runs on a data center GPU or a consumer GPU in an edge cluster. Then the talk goes beyond AI, with GPU-accelerated image processing and scientific compute running as ordinary distributed workloads.

wasmCloud in the Project Pavilion​

You can also find wasmCloud maintainers in the Project Pavilion (Salt Palace, Level 1, Solutions Showcase, Hall 5). It's a great place to learn about the open source foundation of Cosmonic Control, ask technical questions, and connect with the community.

See you in Salt Lake City!​

You don't have to wait until KubeCon to say hello! If you're not already part of the wasmCloud community Slack, come join the conversation. And if you can't make it to Salt Lake City, keep an eye on the Cosmonic blog for updates from the show. We look forward to seeing you there!