Today we are announcing the general availability of EasyStack AI Foundation (EAF AI) — our enterprise-grade platform for Agentic AI. EAF AI gives CIOs and platform teams a single, governed place to build, deploy and operate autonomous AI agents on infrastructure they control.

Over the past eighteen months, every enterprise we talk to faces the same tension: AI pilots multiply faster than governance can keep up. Models live in public SaaS, credentials scatter across notebooks, and nobody can answer the simple question — where does our data actually go?

Why agentic AI needs a new foundation

Classic MLOps platforms were designed for models: train, serve, monitor. Agents are different. They hold long-lived state, call tools, spend money through APIs and collaborate with other agents. Treating them as "just another workload" leaves security and cost holes that only appear in production.

EAF AI starts from the agent lifecycle instead of the model lifecycle. Identity, policy, memory and observability are first-class primitives — not bolt-ons.

What is EAF AI

EAF AI unifies four layers that enterprises previously had to assemble by hand:

  • AI Infrastructure — GPU pooling, fractional scheduling and high-throughput inference serving on EasyStack ECF.
  • Agent Runtime — sandboxed execution with per-agent identity, tool allow-lists and spend limits.
  • Orchestration — multi-agent workflows with human-in-the-loop approvals and replayable audit trails.
  • Governance Console — one pane of glass for policy, model provenance, data residency and cost.
"Enterprises don't need another chatbot. They need an operating system for agents — with the same rigor we apply to core banking infrastructure."

Architecture overview

EAF AI runs on the EasyStack Cloud Foundation (ECF) substrate and inherits its software-defined compute, storage and networking. Agents execute in micro-VMs with hardware isolation, while the AI gateway terminates every model call — applying policy, caching and token accounting before a single packet leaves the platform.

EAF AI reference architecture diagram EAF AI reference architecture: infrastructure, runtime, orchestration and governance layers.

Because the gateway is the single egress point, adding a new model provider — open-source or commercial — is a configuration change, not a re-architecture.

Governance & security by default

Every agent action is signed, attributed and logged. Policy sets answer the questions auditors ask first:

  • Which data classifications may this agent read or write?
  • Which external tools and APIs may it call, and with what spend ceiling?
  • Which actions require human approval before execution?
  • Where must inference run to satisfy data-residency rules?

Combined with ECF's sovereign-cloud deployment modes, this makes EAF AI suitable for regulated industries — government, finance, healthcare and telecom — from day one.

Getting started

EAF AI is available today as part of the EasyStack platform portfolio. Existing ECF customers can enable it through ESCloud Manager with a rolling, zero-downtime upgrade; new customers can start from a three-node reference configuration.

Read the EAF AI whitepaper for the full architecture, or talk to our team to book a technical briefing.

Blog