Guide
What counts as a private AI platform?
A model on your server is just a program that runs. A private AI 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
- Inference
- Open models served privately for chat, coding and embeddings, behind an endpoint your tools can call, with the GPUs and the model lifecycle handled.
- Knowledge
- Your own documents attached to the models through retrieval you configure and can audit. Permissions are respected, so the AI answers from your material rather than the internet.
- Applications
- The apps your workflows run on, with the models behind them.
- Delivery
- How new functionality reaches production: a versioned, audited path from development to production that a person approves.
- Operations
- Monitoring, backups, updates, security, capacity: the layer most offers forget to price. On a platform you run, it belongs to your team or your service provider, and the platform has to make it manageable.
Most offers deliver the first layer, a rented GPU or an installed model runner, and leave the other four as your problem. A platform begins where all five are covered and somebody is accountable for each.
Build or buy?
With an ML engineer, a platform team and a year or more, you can build it; the open building blocks exist. The bill is the people-time and the operations: whoever builds it must also keep it alive, and keep it current as models change. For most companies that does not pay off, which is why a packaged platform exists as a category.
ATAILA as the platform
ATAILA covers the five layers as software on your own hardware, or your service provider’s. It does so in six capabilities, which the platform page describes one by one; their names are enough here:
- Load models
- Model library
- Metering and audit
- AI developer workstations
- Auditable RAG
- Project factory
ATAILA delivers the platform, the migration and third-line support; you or your provider run it. It is built on open source, packaged and auditable. An exit guarantee, written into the contract. The apps you create live in your own repositories.
It is licensed as ATAILA Enterprise for one organisation, or as ATAILA Service Provider for a provider serving its own customers. Priced per CPU socket, not per user or token.
See the platform in detailHow pricing worksOwnership and exit
Which of the five layers is missing in your estate?
Tell us what you run today. We tell you which layers the platform covers for you and which stay with your team.