Tensor9

Tensor9 is an any-prem platform for software vendors. It lets a vendor deploy its full cloud-native stack into a customer’s own VPC or on-prem environment, keeping data in the customer boundary while the vendor keeps centralized control and observability. We assess it because its install-into-customer-env, control-plane, and audited-execution model is a strong pattern for governing AI-driven execution.

Blurb

“Tensor9 is an enterprise any-prem platform. We enable AI vendors to deliver their product inside customer infrastructure.”

Summary

When to use: As a reference architecture for deploying AI and agents into customer-controlled environments, and for any-prem or private-SaaS delivery where data sovereignty is a blocker. Useful wherever AI agents process sensitive data that cannot leave the customer’s boundary.

When to skip: When delivery is purely SaaS and no customer-boundary install is needed. Tensor9 is a deployment platform, not a product on its own.

Key trade-offs:

• Vendor keeps update velocity and centralized control while customer data stays in-boundary
• Control plane lives in vendor’s own account; appliances reach it via outbound-only mTLS tunnel
• Fully logged, authenticated day-2 operations (remote shell, one-off commands, secrets, restarts)
• Complements Kubernetes rather than replacing it

Related garden notes: Agent Client Protocol (registering agent harnesses), Model Context Protocol.

Details

Control Plane Model

Tensor9’s control plane is provisioned in the vendor’s own account. It orchestrates the lifecycle of every customer appliance: compile origin stack to deployable artifact, deploy, then run day-2 operations. Appliances establish an outbound-only, mutual-TLS authenticated tunnel back to the control plane, so there is no inbound network access to customer environments. Every operational action is authenticated via a permissions model and fully logged, giving an audit trail of who did what and when.

Governance and Observability

Tensor9 aggregates metrics, logs, and traces from all distributed deployments and forwards them to existing tools (Datadog, Prometheus). Only metadata leaves customer environments; customer data never touches the control plane. This makes it a template for delivering AI observability, governance, and policy enforcement where the data lives.