Data and AI sovereignty, defined precisely.

• Control

  You decide which data, people, models, and agents can access what, and under which policies.

  Evidence

  • Role-based access control and tenant isolation
  • Data classification and scoped retrieval
  • Model-routing policies by data sensitivity
  • Configurable retention and deletion

• Proof

  You can show what the AI used, what it did, and why, to an auditor, a regulator, or your own board.

  Evidence

  • Source lineage and citations on every output
  • Prompt, tool-call, and action logs
  • Evaluation results before and after release
  • Immutable audit trail for approvals

• Portability

  You are not trapped by one model, one cloud, or one consultant.

  Evidence

  • Exportable data, prompts, and configuration
  • API-first design with documented interfaces
  • Multiple model providers and deployment options
  • Documented handover and exit plan

• PortabilityNIST AI RMF · Govern

  You own the data, prompts, configuration, and outputs

  Everything produced in an engagement is yours, in formats you can use without us.

  • Source data, derived datasets, and embeddings remain in your accounts
  • Prompts, policies, evaluation sets, and agent definitions are delivered as code
  • Outputs, logs, and audit history are exportable at any time
• ControlNIST AI RMF · GovernNIST AI RMF · Map

  No training on your data without written opt-in

  Customer data is not used to train or fine-tune shared models unless you explicitly agree in writing.

  • Model providers are configured with training and retention opt-outs where offered
  • Self-hosted or open-weight models are available where provider terms are unacceptable
  • Any fine-tuning uses your data for your models only, under your agreement
• ControlNIST AI RMF · Manage

  Role-based access and tenant isolation

  Agents and retrieval inherit the permissions of the person or process that invoked them, never more.

  • Permission-aware retrieval mirrored from source-system access rules
  • Scoped credentials per integration and per agent
  • Isolation between business units or customers where required
• ControlNIST AI RMF · Manage

  Encryption, retention, and deletion controls

  Data is encrypted in transit and at rest, kept only as long as you specify, and deletable on request.

  • Encryption using the cloud provider's managed keys or your own
  • Configurable retention windows for prompts, logs, and derived data
  • Documented deletion procedure with confirmation
• ProofNIST AI RMF · Measure

  Audit logs, source lineage, and observability

  Every answer cites its sources. Every action records who triggered it, what the agent saw, and who approved it.

  • Prompt, retrieval, tool-call, and response logging
  • Citations and record-level lineage on outputs
  • Dashboards for quality, latency, cost, and error rates
• PortabilityNIST AI RMF · GovernNIST AI RMF · Manage

  Model-provider and deployment flexibility

  Model choice is a routing policy, not an architecture decision. Switch providers or go self-hosted without a rewrite.

  • Gateway with policy-based routing by data sensitivity, cost, and capability
  • Support for commercial APIs, open-weight models, and private deployments
  • Configurable deployment, data-retention, and model-access options per workflow
• ProofNIST AI RMF · Manage

  Human approval for impactful actions

  Agents draft; people approve. Writes to systems of record, external communications, and financial changes require a named approver.

  • Approval gates defined per action type with escalation and SLAs
  • Reversible actions preferred; irreversible ones require explicit confirmation
  • Approval history retained with the evidence the approver saw
• PortabilityNIST AI RMF · Govern

  Export, portability, and a clear exit plan

  You should be able to leave us, a cloud, or a model provider, and keep the system running.

  • API-first design with documented interfaces
  • Infrastructure and configuration as code in your repositories
  • Handover documentation and a defined exit procedure from day one

• Risk review during Discover covering privacy exposure, unreliable output, bias, and misuse, structured around the NIST AI Risk Management Framework functions: Govern, Map, Measure, Manage
• Evaluation sets built with your team before release; regression checks on every change
• Red-team style testing of prompts, retrieval boundaries, and tool permissions
• Defined incident process: detection, containment, customer notification, root cause, and corrective action
• Data inventory and documented data flows for each workflow, so your privacy and AI obligations can be mapped to concrete controls

DataS AI does not hold SOC 2, HIPAA, PCI DSS, or similar certifications and does not claim legal compliance on your behalf. Our systems are designed to support your governance and compliance requirements (jurisdictional, contractual, and regulatory) by documenting the controls we implement so your program can assess them.

Does our data leave our cloud account?

Only where the workflow requires it and you have approved the path, for example a call to a commercial model API with training and retention opt-outs. Where that is unacceptable, we route to a self-hosted model inside your environment.

Can an agent change data in our ERP or CRM?

Only through a scoped integration and only after a named person approves the specific action. Drafting is automatic; executing is not.

What happens if we stop working with DataS AI?

You keep everything: code, configuration, prompts, evaluation sets, data, and documentation. The exit procedure is defined at the start of the engagement.

Which model providers do you use?

Whichever fit your data sensitivity, cost, and capability needs: commercial APIs, open-weight models, or private deployments. The routing policy is yours to change.