──No Cold Starts

──Scale pods to zero without unscheduling, and wake or migrate them instantly.

──Architect eliminates the tax on idle capacity with no code changes or complicated analytics suites.

Kubernetes scheduling flow diagram

> Tired of Performance Insurance?

Most Kubernetes clusters run with extra capacity to absorb traffic spikes and avoid cold starts. This overprovisioning creates wasted capacity you pay for every hour.

Loophole Labs Architect uses zero-overhead checkpoint/restore on every pod, keeping them scheduled in Kubernetes during sleep. On wake, it eliminates delays from PVC mounts, load balancer registration, scheduling, and software startup, so enterprise and stateful workloads are available immediately.

Architect eliminates the need for overprovisioning Kubernetes clusters and helps you:

Hibernate any Kubernetes pod and wake it without cold starts in under 50ms.

Run pods on spot instances without service disruption on preemption.

Live-migrate stateful workloads across nodes, clusters, or clouds with no service interruption.

# How Architect works

Pod hibernation in Kubernetes sounds simple, but waking a hibernated pod instantly requires complete checkpoint/restore of memory and state, reattachment of storage and network dependencies, and live migration of active connections. Architect does all of this in under 50ms.

Architect is a Kubernetes Availability Platform that handles checkpoint/restore and live migration with a one-line install:

├─1.
Deploy Architect

With a simple change to your deployment and ZERO code modification, you can employ Architect on any cluster for any pod.

├─2.
Continuous checkpoint/restore on every pod

Architect captures pod state efficiently via checkpoint/restore at the containerd shim layer, ready for redeployment or pod hibernation.

├─3.
Hibernate pods, but keep them scheduled

Architect will automatically sleep any pod. Sound dangerous? We give you the ability to tune how Architect behaves based on your needs.

└─4.
Live-migrate or wake pods in under 50ms

On wake, Architect restores complete pod state, reattaches storage, and live-migrates active TCP connections. Pods are ready in under 50ms, no cold start.

● ● ●
loopholelabs — -zsh — 90x25
Acme App
/~\ Architect
# Install /~\ Architect with one command
$ helm install architect architect/architect
# Watch your workloads hibernate automatically
$ kubectl get pods
--------------------------------------------------------------------
NAME STATUS CPU MEMORY
--------------------------------------------------------------------
frontend-7d4b8c6-x9k2 Running 0m 0Mi # <- hibernated
api-server-5f7c9d-h3j7 Running 95m 512Mi # <- active
ml-model-8c9d2f-p4m8 Running 0m 0Mi # <- hibernated
--------------------------------------------------------------------
# First request wakes the pod in <50ms
$ curl frontend.example.com Response: 200 OK (47ms) # <- instant
wake, no cold start
--------------------------------------------------------------------

Architect magic: Pod network migrations without pause and without custom CNI

Architect builds on CRIU and runc checkpoint/restore at the containerd shim layer, with dynamic real-time routing that live-migrates active TCP connections via eBPF. No custom CNI plugins, no complex setup, universal compatibility with any Kubernetes cluster.

Compatible with Tigera, Cilium, and AWS VPC CNI

! Let Your Cattle Sleep Standing Up

Manually hibernating a pod requires code changes and manual state offload. Architect automates checkpoint/restore so you do not have to.

OLD WAY ~ MANUAL /

Code modification required.
Manual offload of state to external storage.
Checkpointing limited to partial pod state.
Data loss on node failure.
Startup time in seconds.
Network migration with custom CNI and not instant.

NEW WAY ~ WITH ARCHITECT /

No code changes. One-line install.
Automated checkpoint/restore at the containerd shim.
Full image integrity checked.
No data loss on node failure.
Startup in as little as 50ms.
Live migration of IP traffic via eBPF.

/ Works the way you work

Architect runs on any standard Kubernetes configuration with no code changes and no custom CNI. Multi-cloud and multi-cluster compatible. Try it today via our quick start.

Save as you scale

Hibernate idle Kubernetes pods to scale to zero. Better bin-packing, lower compute cost.

No cold start, ever

Pods wake in under 50ms. Even complex enterprise and stateful workloads.

Kubernetes native

Architect follows the same patterns and exposes primitives the way you already work.

Architect integrates with Valkey, Kafka, PyTorch, vLLM, NVIDIA and deploys to any cloud

> Get started: Run the way you run today

Architect installs in one helm command. No code changes, no analytics tooling, no custom CNI. Label your pods and you are ready.

󰆾

Build your containers with standard tooling. Ship them without code modification.

󰦕

Deploy with a helm chart on any Kubernetes cluster, including spot instances across nearly any cloud.

󰆾

Hibernate idle Kubernetes pods or live-migrate them anywhere, optimizing cost and utilization without interruption.

   AAA    RRRRRRRR   CCCCCC  HH     HH IIIIII TTTTTTTT EEEEEEE  CCCCCC  TTTTTTTT
  AA AA   RR     RR CC    CC HH     HH   II      TT    EE      CC    CC    TT
 AA   AA  RR     RR CC       HH     HH   II      TT    EE      CC          TT
AA     AA RRRRRRRR  CC       HHHHHHHHH   II      TT    EEEEEE  CC          TT
AAAAAAAAA RR   RR   CC       HH     HH   II      TT    EE      CC          TT
AA     AA RR    RR  CC    CC HH     HH   II      TT    EE      CC    CC    TT
AA     AA RR     RR  CCCCCC  HH     HH IIIIII    TT    EEEEEEE  CCCCCC     TT