AI and ML
Was mich Dutzende KI-Agenten gelehrt haben: Wie ich das Speichersystem Blockstor als Experiment geschrieben habe
Andrei Kvapil über Blockstor: ein Clean-Room-Orchestrator für Blockspeicher, gebaut mit bis zu 60 KI-Agenten, testgetriebener Entwicklung und harten Exit Gates.
What dozens of AI agents taught me: how I wrote the Blockstor storage system as an experiment
Andrei Kvapil on building Blockstor as a clean-room, Kubernetes-native block storage orchestrator by driving up to 60 AI agents with TDD and hard exit gates.
VMware replacement after Broadcom: a guide for service providers, banks, and sovereign clouds in 2026
What changed under Broadcom, a component-by-component VMware-to-Cozystack mapping, how the migration actually runs, and the FAQ engineers ask first.
Transport and logistics cloud architecture — NIS2, AI, edge in 2026
A three-tier architecture for transport and logistics, the NIS2 controls that apply to the sector, and where AI workloads fit.
Telco cloud modernization in 2026 — from legacy NFV to Kubernetes-native edge
How tier-1 and tier-2 telecom operators modernize legacy NFV environments into Kubernetes-native sovereign cloud platforms — a guide for architects.
Telco-Cloud-Modernisierung 2026 — von Legacy-NFV zur Kubernetes-nativen Edge
Wie Tier-1- und Tier-2-Telcos Legacy-NFV-Umgebungen zu Kubernetes-nativen, souveränen Cloud-Plattformen modernisieren — ein Leitfaden für Architekten.
Seven decisions when designing sovereign AI architecture
Seven architecture decisions behind a sovereign AI stack, how they interlock, and the combinations that recur in real deployments.
Sieben Entscheidungen beim Entwurf einer Sovereign-AI-Architektur
Sieben Architekturentscheidungen hinter einem souveränen AI-Stack, wie sie ineinandergreifen und welche Kombinationen in realen Deployments immer wiederkehren.
Reverse cloud migration — a practical playbook for leaving public cloud in 2026
A five-step cloud repatriation playbook, the pitfalls that recur, when not to repatriate, and how long a realistic move actually takes.
Smart grid platform architecture — IT/OT convergence, edge, and AI on customer-controlled infrastructure
A smart-grid architectural reference for energy operators: IT/OT boundaries that work, NIS2 controls, AI on grid-operational data, and legacy migration.
Private LLM deployment — a practical guide to on-premise AI infrastructure in 2026
The six layers of a real private LLM deployment — hardware, platform, serving, model, application, operations — with the pitfalls at each one.
Ein Cloud-Produkt für Kunden starten — Playbook für Hosting-Anbieter, Telcos und regionale Betreiber
Die sechs Schichten eines Cloud-Produkts für Endkunden, die Architekturentscheidungen einer Public Cloud und woran der Markteintritt kommerziell scheitert.
Industrie-4.0-Plattform — Cloud- und Edge-Architektur für die Fertigung 2026
Industrie-4.0-Architektur 2026: Muster von der Edge bis zum Core, Souveränität für Industrie-IP und die NIS2-Pflichten, unter die Hersteller jetzt fallen.
Industry 4.0 platform — cloud + edge architecture for manufacturing in 2026
Industry 4.0 architecture in 2026: edge-to-core patterns, sovereignty for industrial IP, and the NIS2 controls manufacturers are now in scope for.
Launch a customer-facing cloud product — playbook for hosting providers, telcos, and regional operators
The six layers of a customer-facing cloud product, the architectural decisions specific to public cloud, and where launches stumble commercially.
K-12 school district cloud infrastructure — when sovereignty matters more than convenience
Why K-12 infrastructure differs from universities, when a district actually needs sovereign infrastructure, and the architecture pattern that fits.
Hybrid cloud architecture patterns 2026 — what works, what fails, and how to choose
Five hybrid cloud patterns that work in production, what makes them work, the failure modes to avoid, and when hybrid is the wrong answer.
Hosting provider platform modernization — from VPS to cloud product
Architectural starting point, migration sequencing, and unit economics for hosting providers modernizing onto a Kubernetes-native multi-tenant platform.
DevOps best practices for 2026 — beyond the slide-deck era
The eight DevOps practices that compound in 2026, what is still contested, the failure modes that recur, and how to place your team on the maturity curve.
Cloud-native Infrastruktur für Forschung und Lehre — was Hochschulen 2026 wirklich brauchen
Architekturmuster für Forschungs- und Lehrinfrastruktur an Hochschulen: die drei Aufgaben, GPU-Scheduling für Labore und die wiederkehrenden Fallstricke.
Cloud-native research and teaching infrastructure — what universities actually need in 2026
Architecture patterns for university research and teaching infrastructure: the three missions, GPU scheduling for labs, and the pitfalls that recur.
AI Platform — wann sich Dauer-Inferenz auf eigenen GPUs gegenüber Hyperscalern rechnet
GPU-Wirtschaftlichkeit für Dauer-Inferenz, mandantenfähiges GPU-Scheduling und ab wann sich eine eigene KI-Infrastruktur gegenüber Hyperscalern auszahlt.
AI Platform — when sustained-inference economics beat hyperscaler GPU
GPU economics for sustained inference on the Ænix AI Platform: multi-tenant GPU scheduling, fractional sharing, and when dedicated AI infrastructure pays back.
Cozystack v0.35: External Application Sources, Dedicated S3 Clusters and Hetzner RobotLB Support
Cozystack v0.35 adds external application sources for its modular architecture, dedicated S3 clusters with monitoring and Hetzner RobotLB support.
Invitation to CozySummit Virtual — December 3
Join us on December 3 for CozySummit Virtual, the first conference for CozyStack developers and users.
Cozystack v0.31–0.33: Air Gap, Backup System, AI Workloads in Kubernetes and More
Cozystack v0.31 to v0.33 bring air-gapped installs, a new backup system, AI workloads in Kubernetes, ARM support, NFS and cozypkg as a Helm replacement.
Cozystack Recognized in the CNCF Cloud Native AI Landscape
The Cloud Native Computing Foundation has added Cozystack to its Cloud Native AI (CNAI) Landscape as a platform for running AI workloads on Kubernetes.
Cozystack Now Offers GPU Passthrough for AI/ML Virtual Machines
Cozystack now passes GPUs directly through to KubeVirt virtual machines, so tenants can run AI, machine learning and other GPU workloads in their VMs.
Ænix Secures $300K Seed Investment from Prospective Technologies to Accelerate Open-Source Cloud Innovation and the Cozystack Platform
Ænix, the company that created the open-source Cozystack platform, announces a $300,000 seed investment from Prospective Technologies (April 2025).