SOVATATrust, Security & AI Governance
Sovata earns enterprise trust through architecture, not assurances. Security, data ownership, governance and privacy are designed into every private AI deployment from the first boundary conversation — inside infrastructure you govern, with no vendor lock-in and a human accountable for every consequential decision.
Six commitments, built into the system
Each of these is a design decision made before your deployment goes live — and monitored after it does.
Security-first architecture
Security is designed in from the first boundary conversation, not bolted on after go-live.
Access control, data-flow rules, PII handling and audit logging are defined during the “set the boundaries” step — before a line of code is written. The deployment works within your existing security and compliance framework rather than asking you to adopt a new one.
- Least-privilege access and role-based controls
- Explicit data-flow boundaries — nothing leaves without a rule allowing it
- Guardrails for PII, safety and validation at the orchestration layer
- Full audit trail for oversight and review
Data ownership
Your data — and everything derived from it — stays yours. We don’t train on it or reuse it elsewhere.
The instance runs inside your environment. Your knowledge, prompts and outputs are processed there, not sent to an external provider, and are never used to build products for anyone else. Ownership of the data and the decisions about it stays with your organisation.
- Processing stays inside your governed environment
- No training on your data for third parties
- You own the knowledge base, the configuration and the outputs
- Data residency (e.g. onshore NZ) agreed up front where required
No vendor lock-in
Built so you can change model, provider or direction without a rebuild — and without us holding the keys.
The architecture separates your knowledge and configuration from any single model. Because it lives in infrastructure you govern, you’re never trapped by one vendor’s pricing, roadmap or deprecation schedule — the risk that breaks so many cloud-AI automations.
- Model-agnostic orchestration — swap models without rebuilding
- Your configuration and knowledge are portable and documented
- Runs in infrastructure you control, not a proprietary black box
- Predictable, fixed running cost instead of per-query metering
AI governance
Explicit rules for what the AI can do, must escalate, and must never do — agreed before deployment.
Governance is a first-class part of the build: what the system may access, when it must hand off to a human, and the limits it can never cross are defined with your team and monitored after go-live. This is the difference between a governed production system and an ungoverned experiment.
- Access, escalation and “never do” rules agreed with your team
- Human accountable for every consequential decision
- Ongoing monitoring, not ship-and-forget
- Aligned to responsible-AI practice and your regulatory context
Privacy commitments
Privacy by design: the system only sees what it needs, and only surfaces what it’s allowed to.
Boundaries reflect your obligations and, where relevant, your tikanga — including where certain knowledge should not be surfaced at all. We design those limits with you, not impose a template, and we’re explicit about what the system can and can’t do.
- Data minimisation — the model only accesses what the use case needs
- Culturally- and policy-aware access boundaries
- Clear, plain-language record of what the system does with data
- Designed to support your privacy and sovereignty obligations
Enterprise-ready infrastructure
Operated, monitored and improved after launch — a production system, not a proof of concept.
A private AI instance is operated, not shipped and forgotten. We onboard your team, monitor performance, keep governance current, and improve the system as your needs change — with senior people who built it staying involved well beyond go-live.
- Monitoring, performance tuning and continuous improvement
- Senior team stays involved after launch — no hand-off
- Scales with your usage on predictable infrastructure
- Documentation and knowledge transfer built in
Discovery → Design → Build → Deploy → Support
A structured, low-risk path from AI uncertainty to a governed production system — with governance sequenced before build so your stakeholders are aligned before the system exists.
- Discovery01
Understand the problem
A free 30-minute conversation to find the work where private AI gives the most value with the least risk — and an honest read on whether it’s a fit at all.
- Design02
Set the boundaries
Data rules, access controls, escalation points and what the AI must never do — agreed with your stakeholders before any technical build begins.
- Build03
Prepare the knowledge
Structure, test and validate the content the AI will use, inside your environment, so it answers accurately within approved boundaries.
- Deploy04
Launch and onboard
Go live inside your infrastructure, onboard your team, and turn governance from a design document into live, monitored controls.
- Support05
Operate and improve
Ongoing monitoring, governance upkeep and improvement as your needs change — the senior people who built it stay involved.
New company, senior experience, honest scoping
New Zealand-based, New Zealand-first
Sovata is based in Auckland and built for New Zealand’s sovereignty context — from Privacy Act 2020 obligations to Māori data sovereignty — with data residency agreed before any build begins.
Senior people, no hand-off
The people you meet in discovery are the people who design, build and run your system. You don’t get sold by seniors and delivered by juniors.
Real enterprise track record
Ex-IBM enterprise technology, 16 years in banking data governance at Kiwibank, a seat on the AI Forum NZ Pacific Advisory Panel, and governed AI delivered for healthcare triage.
Honest scoping
Every engagement starts with a free discovery call, and if a private deployment isn’t the right answer for your use case, we tell you.
Proof you can inspect
We’ll show you what we’ve built — including full products we’ve designed and shipped — and be candid about what we can and can’t do.
Trust & security questions
No. Your data is processed inside your environment for your use case only. We don’t train shared or third-party models on it, and nothing derived from it is reused to build products for anyone else.
Because the deployment lives in infrastructure you govern and your knowledge and configuration are portable and documented, you keep the system. There’s no proprietary black box holding your data or your operations hostage — that’s the point of avoiding vendor lock-in.
If onshore residency is a requirement, it’s designed in during the boundary-setting step before any build begins — not retrofitted. We agree data residency with you up front.
Boundaries are defined per engagement against your regulatory context, existing infrastructure and internal policies. We work within your organisation’s existing security and compliance framework rather than asking you to adopt a new one.
Your organisation stays accountable for consequential decisions, by design. Governance defines what the system may do alone, what it must escalate to a human, and what it must never do — agreed before deployment and monitored after.
Ready to talk it through?
Book a free discovery call. No preparation required — just tell us what you’re trying to solve.