One catalog for every AI system

Why this conversation

Large organisations adopt AI fast, because the technology moves fast. Many teams pick many tools, and before long nobody can see the whole picture: which models are in use, who has access, and what each system is for. In Europe, the EU AI Act adds a compliance question on top. On 18 September, at AGNTCon + MCPCon Europe in Amsterdam, the host spoke with Roman Swoszowski, VP of Product Development at Grape Up, about Aiboostr. The rest of the week is in AGNTCon + MCPCon Europe 2026: who we met.

Grape Up has been on the market for about twenty years as a service and technology consulting company, he says, and is now shifting towards a product-based approach. Aiboostr, an AI governance and orchestration platform for enterprises, is its product.

One catalogue, and access on request

Aiboostr gives an organisation a single control plane. At its centre is an inventory of all its AI systems, agents and models, public and external as well as private, including any local LLMs, in one catalogue. The host asked if that was a registry; Roman agreed it is.

With the catalogue in place, an end user can request access to a model or a tool and have it approved straight away. That speeds up adoption inside the organisation while keeping it controlled, because the company knows who uses what.

The EU AI Act as a side effect

Compliance with the EU AI Act, he says, comes as a side effect of having the catalogue. Each AI system can carry attributes: its risk classification, intended users, purpose, whether it has been reviewed and whether it is approved for use. Attaching those to a model or a tool becomes easy, and the catalogue can also act as a data source for an external GRC system that holds a company’s risk and compliance information.

On-prem, from proof of concept to production

Aiboostr is offered primarily on-prem, deployed in the customer’s own infrastructure. It is not a SaaS tool, because of data privacy and data residency: customers typically prefer to keep all of it in-house. Engagements start with a proof of concept, a dedicated simple deployment integrated with some of the tools or agents the customer already has, which is later promoted to a full production environment. It runs natively on Kubernetes, so any existing cluster will do, OpenShift and Rancher included.