Guide

What counts as a private AI platform?

A model running on your server is not a platform — it's a running program. A platform is what your company can work on: models, knowledge, applications, delivery and operations, together. This page lays out the five layers — and where offers usually fall short.

The five layers

1. Inference

Open-weight models served privately (chat, coding, embeddings) behind an OpenAI-compatible API — with GPUs, scheduling and model lifecycle handled.

2. Knowledge

A private knowledge base (RAG) built from company documents: indexing, permissions, refresh — so the AI answers from your material.

3. Applications

The apps your workflows actually run on: web, database, storage, email, integrations — because a model alone is not a workflow.

4. Delivery

How new functionality reaches production: a versioned, audited, human-approved release path (DEV → UAT → PROD).

5. Operations

Monitoring, backups, updates, security, capacity — the layer most offers forget to price.

Most offers on the market deliver layer 1 (a rented GPU, an installed model runner) and leave layers 2–5 as "your problem". A platform begins where all five are somebody's contractual responsibility.

Build or buy?

Honestly: with an ML engineer, a platform team and 12+ months, you can build it — the open components (vLLM, Open WebUI, Postgres, MinIO, Kubernetes) exist. The bill isn't the software; it's the people-time and the operations: whoever builds it must also keep it alive. For a 20–200-person company that rarely pays off — which is why the operated platform exists as a category.

ATAILA as the operated platform

All five layers, one contract: private inference on EU GPUs · a knowledge base from your documents · the full application stack · an audited, human-gated release path · and the team that operates it — under a written 99.9% SLA, for a flat monthly fee. Built on open components with a documented export: you can leave any time (ownership →).

From €500 / month (net anchor), operated. How pricing works →

Curious what it means for your case?

Tell us the workflow — we'll show you the platform running it.

Talk to us →