01 / Govern access
Make identity and data class part of every route.
Evaluate the user or application, allowed profile, trust boundary, and active policy before cost or latency can influence a decision.
Enterprise
Enterprise AI control across models, providers, and applications.
Tensor Cortex gives platform, security, and engineering teams one governed contract for model access. Every evaluation starts with one real route, its policy boundary, and the evidence needed to accept it.
Source reviewed:
Current evaluation boundary
- Product
- Private release candidate
- Access
- No public production API or self-service signup
- Review
- Deployment-scoped and approved case by case
Control outcomes
A stable access layer for a changing model market.
The product contract keeps policy ahead of provider selection, so teams can evaluate model access as an operating system rather than a collection of disconnected proxies and keys.
02 / Stabilize integration
Expose logical capabilities instead of provider churn.
Applications target an OpenAI-compatible contract and logical profiles. Physical endpoints can change only inside the approved deployment and compatibility boundary.
03 / Inspect decisions
Join route, attempt, usage, and audit evidence.
Request identifiers and content-negative records make policy and fallback behavior reviewable without treating prompt collection as the default.
Operating models
One product contract. Three deliberately distinct states.
A delivery model is not available merely because its architecture exists. The current status of each model remains visible throughout evaluation.
Cloud / Shared
Invite-only alpha preparation
The planned Tensor Cortex-operated shared alpha is limited to at most ten organizations and public-provider BYOK. It is not an active public service and excludes private endpoints in this scope.
Self-Managed / Single tenant
Current evaluation direction
Customer-dedicated release-candidate evaluations may be scoped and approved case by case. Evaluation does not imply production readiness, an SLA, or general availability.
Dedicated / Managed
Planned, not available
A Tensor Cortex-managed single-tenant model remains planned. Repeatable managed operations and external release evidence are not yet complete.
Evaluation path
Start with a testable decision, not a broad rollout.
The shortest useful evaluation isolates one buyer-critical route, documents the boundary, and decides against written evidence.
- Define the route Identify the user or application, data class, model capability, and trust boundary.
- Agree the evidence Choose the policy, compatibility, failure, audit, and usage outcomes that must be demonstrated.
- Use an approved environment Bind credentials, providers, regions, retention, and operating responsibility to the evaluation.
- Close in writing Record acceptance, gaps, commercial terms, and the next stage without converting a test into an SLA.
Buyer packet
Review the public boundary before requesting private evidence.
Product behavior, security architecture, legal terms, and company identity each have an authoritative public surface.
Architecture
Security behavior
Inspect policy-first routing, content-negative audit behavior, authentication, streaming, and safe error contracts.
Review the security architecture →Contract
Gateway interface
Inspect authentication, streaming, errors, route metadata, and compatibility limits.
Read the API reference →Legal
Website privacy and terms
Separate public-site handling from the agreement required for a product evaluation.
Read the privacy policy →Company
Buyer-facing company facts
Confirm the legal entity, jurisdiction, operating principles, and company contact.
Review company information →Bring one real control problem.
Request a deployment-scoped review. We will assess fit, identify the required environment and credentials, and propose the next evidence step for one buyer-critical route.