AI · Customer-controlled Data Mediation · NVIDIA OpenShell
Introducing TomorrowX Data Mediation™ for NVIDIA OpenShell.
Put agentic AI to work on your terms. AI agents become valuable when they can use the systems, tools and information your organisation already depends on. TomorrowX helps you govern those interactions, protect sensitive information, apply your own model and data policies, and retain evidence of the decisions made, all within an architecture you control.
Data Mediation complements NVIDIA OpenShell’s secure agent runtime by adding independent, programmable controls to the data exchanges selected for mediation. The controls are shaped by your workflows, operating environment and risk requirements.
OpenShell governs whether an agent can act. Data Mediation governs what data moves when it does.
Authorities in Australia, India, Singapore, the United Kingdom and the United States are paying increasing attention to how organisations give AI systems access to data, tools and real-world actions. The common themes are clear: bounded access, strong identity, continuous observation, accountable decision-making and controls beyond model instructions.
For customers, this is an opportunity to move forward. When the organisation can define the boundaries and demonstrate that they are respected, Risk and Cyber can help enable more valuable AI use cases, not just assess them after the fact.
Guidance and advisories are provided as context. They do not endorse TomorrowX or mandate a particular control architecture.
One solution. Four essential capabilities.
Independent data controls for the interactions you choose to mediate.
01 — Protect the information your organisation is responsible for
Apply policy to selected mediated exchanges. Mask, minimise or withhold sensitive data where configured, so agents can perform authorised work without exposing personal, commercial or regulated information.
02 — Use the model that fits the task and your policies
Route eligible workloads according to policy, capability, destination, sensitivity, residency and cost, across private, sovereign and cloud models.
03 — Connect agents to the systems they need, on approved terms
Govern selected requests and responses between agents, models, APIs, tools and enterprise systems, without forcing the underlying systems to be rebuilt.
04 — Show what happened, what was permitted and why
Collect logs and evidence from mediated exchanges, and complement them with OpenShell and application audit sources, to support security teams, risk functions and ongoing assurance.
How Data Mediation works with OpenShell
Your requirements determine what needs to be controlled.
Some organisations need to protect sensitive information before it reaches an external model. Others need to govern which tools agents may use, limit what is shared with connected systems, or retain evidence of decisions and exchanges. We start with those requirements and establish the appropriate control points in your environment.
Independent controls can be applied at approved points in the data paths your organisation chooses to mediate. Controls act on selected, routed exchanges; they do not read an agent’s internal state, local files or traffic that is not routed through them.
Technical note · placement patterns
For architects: where Data Mediation can be placed.
Three placement patterns, chosen to fit the customer’s environment, security requirements, use cases and OpenShell version. Further placements, such as mediation beside an egress gateway for non-agent traffic and log shipping, are added as requirements identify them.
1 · Model endpoint mediation
Data Mediation operates as the policy-approved model endpoint the sandbox is allowed to reach, enabling sensitive-data protection, intelligent model selection and controlled use of local or external AI models.
2 · Supervisor middleware
TomorrowX’s multi-protocol Programmable Data Agent connects through OpenShell’s documented gRPC supervisor middleware interface, bringing content-aware policy enforcement to policy-matched agent requests and responses.
3 · Agent ingress mediation
Data Mediation can also be deployed in front of agent-facing interfaces, governing information received from users, applications and other agents before it reaches the sandbox.
Deploy within your environment
Retain control.
Data Mediation is licensed software deployed inside customer-controlled infrastructure.
No requirement to migrate enterprise data
No requirement to modify existing applications
No requirement to send customer data to TomorrowX
Supports cloud, hybrid, on-premises and sovereign architectures
Designed to work across modern and legacy technologies
Connect, don’t integrate. Control without compromise.
Put agentic AI to work
Start with a workflow worth improving.
Bring us the use case, the systems it needs to reach and the controls your organisation cannot compromise. We will work with your team to define the boundaries, demonstrate the relevant Data Mediation capabilities and agree what successful adoption would look like in your environment.
Regulatory references are provided as industry context. They do not constitute an endorsement of TomorrowX. Data Mediation for NVIDIA OpenShell is a TomorrowX solution; it is not a joint NVIDIA product, certification or endorsement. TomorrowX is a member of NVIDIA Inception.