Notizen aus der Plattform.

LLMOps-Observability verbindet Infrastruktur, Runtime und Analyse, damit Teams Änderungen bei Latenz, Kosten und Qualität auf ihre Ursache zurückführen.

Anleitung: Code von KI-Agenten in Kata-Containers-VMs auf Amazon EKS ausführen, verwaltet mit Agent Sandbox, ohne Egress und ohne Zugangsdaten.

When you start using a cloud provider in an enterprise scenario, with multiple teams, multiple environments and real data, you quickly find out that the hard part is not picking the right instance type or the correct database. The hard part is the infrastructure underneath: who has access to what, how networks talk to each other, where logs go, how secrets are stored. If you sort these things out before deploying your first application rather than after, you save yourself a lot of future headaches. The concept of landing zone is exactly this: a pre-configured and standardized foundation, upon which applications are built. Instead of letting application teams figure out solutions for networking, logging, and access management on their own, platform teams provision a common set of resources that address those needs before any application is deployed. For this reason, defining a landing zone is the recommended first step in any organization's cloud adoption process. At Liquid Reply, we design and build multi-cloud enterprise-ready solutions, combining security principles, Infrastructure as Code, and platform engineering practices to create secure, repeatable and operationally manageable environments for application teams. In this series of blog posts, we will show how this knowledge can be applied to the creation of a landing zone for Scaleway. Scaleway is a French cloud provider that is increasingly used by European organizations, especially by those that require data sovereignty. This post covers the reasoning behind the design, the Scaleway services involved, and a complete architectural view. Later posts will go deeper into the implementation and possible future improvements.

Ein Follow-up zu unserem Artikel im AAIF-Blog „Designing RequestState for Multi-Round Trip Requests“. Der gesamte Code in diesem Beitrag stammt aus dem Begleit-Repository `openbao-mcp-requeststate` und läuft gegen ein echtes OpenBao und das offizielle MCP Python SDK.