Pradeep Dahiya
Founder and Architect, ScotiTech Solutions Limited
Regulated teams have been told for two years that they must choose between modern AI and control of their own data. That is a false choice, and it costs firms a lot of avoidable procurement time. This guide gives a working decision framework — the questions we actually ask on a discovery call.
The false choice
Most AI vendors sit on one of two extremes. On one side, the multi-tenant SaaS assistant with impressive product design and a data pipeline pointing at their cloud. On the other, a boutique on-premises consultancy that will happily build something bespoke over the next twelve months.
A regulated buyer wants neither. They want AI capability that behaves like the rest of their software estate — deployed inside their boundary, run by their operations team, and reviewable by their security team.
What "self-hosted" actually means for AI
Self-hosted AI is not one product; it is a deployment posture. The application runs inside customer infrastructure. The database sits in the customer environment. The retrieval index is customer-owned. The LLM call routes to a model endpoint the customer picked and paid for.
This posture is what makes AI adoptable in law firms, hospitals, financial services, and manufacturing operations. It is not new — it is the same posture your ERP, SIEM, and identity provider already sit in.
Where SaaS still fits
For unregulated content and internal experiments, SaaS AI is still the fastest way to try things. It is only when customer-material, patient-adjacent, or supervisory-board content enters the picture that the SaaS pattern hits a wall.
Five questions that decide the answer
When we run a discovery call with a new team, five questions do most of the work:
- Does the workload include material your compliance function cannot see leaving the firm — client documents, patient-adjacent records, supervisory reporting?
- Are you already running an SIEM, an identity provider, and a change-controlled deployment pipeline? If yes, the operational cost of self-hosting is low.
- Do you have a strong preference for a specific LLM provider, or are you comfortable being told which one you must use?
- Do you need the answer to be reproducible against the same source content later — for audit or regulatory review?
- Is a per-seat SaaS contract a growth penalty for you at your current headcount trajectory?
Where AXOS sits
AXOS is designed for teams where at least three of the five answers point at self-hosted. The platform ships as a small set of container images that run inside the customer environment, connects to the customer identity provider, and lets the customer choose which model endpoint receives the prompt. There is no AXOS-operated cloud in the production data path.