Hardware-Isolated Containers,
at Native Speed
Edera Hardened Runtime runs each Kubernetes pod with its own kernel for true hardware isolation at performance within 5% of native.
.png)
How It’s Done
Drop it into any Kubernetes cluster as a RuntimeClass and run untrusted workloads, multi-tenant pods, and AI agents with containment, by design – no rearchitecting required.
Edera Platform
Ships with Your Stack
THE PROBLEM
The Shared Kernel is the Blast Radius
The threat model has shifted: AI-scale vulnerability discovery surfaces novel zero-days in the kernel in hours, not years, so "zero CVEs" no longer means "zero exploitable bugs." With GPUs and untrusted AI agents handling sensitive data on shared nodes, the kernel boundary is now the line between your workloads and everyone else's.
.png)
How It Works
A Hardware Boundary Around Every Pod
Edera Hardened Runtime puts a true hardware-isolation boundary around every pod and runs each one in its own microVM with its own Linux kernel, called a zone. With no shared kernel to escape, each workload is contained in hardware rather than software policy that has to be configured for every workload.
Install as a RuntimeClass
Deploy Edera and point your workloads at it with a standard Kubernetes runtimeclass. No control-plane, node, OS, or app changes.
Each Pod Gets a Zone
Every pod boots into its own microVM with a dedicated Linux kernel, isolated at the hardware boundary from the host and from every other pod.
Your Images Run As-Is
Because each zone runs a complete Linux kernel, your existing container images, syscalls, and tooling just work, no WebAssembly, no rebuilds, no surprises.
Operate it Like Any Pod
Zones schedule, autoscale, and report metrics like normal pods. kubectl top, HPA, and your dashboards all work at performance within 5% of native.


Early Access
Edera KVM early access is open now: try it on your own stack and help shape what GA looks like.
Core Features
Why Edera Is Different
Every pod runs in its own microVM with its own Linux kernel. Remove the shared kernel and you remove the lateral movement, privilege escalation, and node-wide blast radius that come with it.
Workloads run within 5% of native performance, and in a customer's production environment Edera pods ran ~20% faster and up to 40% more efficient than Kata.
Runs on Kubernetes, EKS, and more — across Amazon Linuz, Ubuntu, and other OSes — with no control-plane, node, OS, or image changes. It just works.
Run any instance, any public cloud, on-prem – your infra, your rules. You get the full Edera platform running on infrastructure you control.
Pin a different kernel and driver version per zone on shared GPU nodes. Isolate tenants and let platform teams match driver stacks to workloads — without the conflicts and passthrough fragility of microVM GPU stacks.
kubectl top, horizontal pod autoscaling, and your existing metrics just work. Zones report like normal pods, so you keep the visibility that microVM stacks usually break.
DEEP DIVE
Where Edera Sits In Your Stack
.png)
Secure Infra, Lower Costs
Single-tenant burns budget on idle capacity; multi-tenant shares a kernel with real containment gaps. Edera gives every workload its own lightweight VM – one Fortune 500 cut footprint 62%, saving millions.
Ship Agents, Capture Return
90% of agent pilots stall in review – there's no mechanism to contain what an autonomous agent might do. Edera's hardware-isolated VM per agent makes containment real, so agents ship and revenue lands.
Remove the Risk
Minimal images and eBPF monitoring help reduce the attack surface, but containers still share one kernel – a bug can cross that boundary. Edera gives every workload its own kernel, removing the risk entirely.
Try One Node of Edera Today
You're running untrusted workloads on infrastructure built to share everything. Every AI agent, every model execution, and every third-party container is a shared kernel away from your host. Edera closes that gap–without the compatibility limitations, overhead, or hardware dependencies of existing alternatives.
USE CASES
Built For Your Untrusted Workloads
Give an autonomous agent its own sandbox – a VM with its own kernel – to run untrusted, generated code in. It can do whatever it needs while the host and neighboring workloads stay out of reach.
Run untrusted tenants on shared nodes without a shared kernel. Contain every tenant to its own zone so one compromise can't reach another.
Execute customer-supplied builds, plugins, and CI jobs in disposable, sandboxed zones with dedicated kernels — no lateral movement, no node-wide blast radius.
Share GPU nodes across tenants with per-zone kernels and drivers, ending the driver-conflict and passthrough fragility of legacy microVM stacks.
Hardware-enforced, per-workload isolation gives regulated organizations a straightforward path to satisfying NIST SP 800-190 container security controls, NIST 800-53, and FedRAMP High/DoD baseline requirements.
Edera supports confidential computing by providing strong architectural isolation that bridges the gap between hardware-bound TEEs and modern container workloads.
Resources
A Curated Collection of Musings & Research

YOU KNOW YOU WANNA







.png)
.png)
.png)


