Managed
We operate it. Fastest route to a first deployment, suitable where the data classification allows it.
StromaOS runs in four deployment postures, differing in exactly one respect: who operates the compute. In your VPC, embedded in your data platform, or air-gapped, nothing in the request path is ours to host. Managed is the exception - we operate it, under the data classification your agreement sets. The governance envelope is identical in all four.
Commitments you can verify during an architecture review rather than take on trust.
The deployment model changes where the work runs. It does not change what is enforced.
We operate it. Fastest route to a first deployment, suitable where the data classification allows it.
Runs in your cloud account, inside your network boundary, under your identity provider and your controls.
Deployed natively into the data platform you already run, so work executes where the data already lives.
Fully sovereign, no egress, your own models. For classified programs and jurisdictions that require it.
Mapping, de-identification and entitlement run as governed steps against your sources rather than by migrating your data into a vendor environment and selling it back to you. In self-operated postures those steps execute entirely within your boundary; under managed deployment they execute within the boundary your agreement defines.
Semantic transformation maps your sources to a common model where they sit. Systems of record are not replaced and data is not migrated to us.
PHI and PII removal runs as a governed step at the point of preparation - not as a preprocessing task someone remembers to do - with a record of what was protected and how attached to the run.
Licensed assets resolve only when your entitlement verifies. A run that loses entitlement withholds its result rather than discarding the work.
A log tells you what happened. A constraint tells you what can happen. These are behaviors you can observe during an architecture review.
Policy is declared as code, bound to the work, and evaluated against live context - which user, which tenant, which study, which data classification - before an action reaches a model.
A control that cannot be evaluated stops the run. A check that cannot reach its service refuses; it does not pass. Unknown is never recorded as none.
Selection resolves to a named authority inside your organization, recorded with who signed and what the signature meant. There is no path where the platform decides and you inherit the liability.
Abrupt termination leaves half-written batch records. New dispatch halts, in-flight work completes to the next transactional boundary, and resumption requires explicit approval.
Every enterprise buyer asks this, and in regulated pharma they ask it early.
Three of the four postures put execution on your compute inside your own boundary, where your existing controls apply rather than being duplicated. Managed exists for speed where the data classification allows it.
Bring your own. Nothing is required, resold or metered, and we do not compete with the providers you have standardized on.
Open schema, queryable, exportable. The retention obligation on regulated evidence outlives any vendor relationship, so the record has to be readable without us.
An architecture review maps our governance artifacts to your evidence model, and shows you the gaps in both directions.