ZTA Labs

Open ecosystem for institutional investors

Designed for an open institutional ecosystem.

ZTA Labs is designed for asset managers and asset owners that already have established data, infrastructure, applications, security controls, and service relationships. The open, deployable reference architecture is designed to fit that environment rather than force a rip-and-replace decision.

How we think about partners

The architecture should integrate with the institution’s operating environment.

ZTA works across an open partner ecosystem that can include enterprise data platforms, open-source software, open-weight model providers, infrastructure providers, identity and security systems, workflow platforms, and implementation partners.

Partner strategy follows the client’s environment and governance requirements, not the other way around.

Enterprise data & content

Platforms that hold or govern institutional evidence, entitlements, and metadata.

Inference & model operations

Serving, routing, evaluation, and orchestration capabilities for approved open-weight models.

Infrastructure

NVIDIA infrastructure and the software layers required to deploy client-controlled AI environments.

Identity & security

Institutional identity, access control, audit, secrets, encryption, and operating controls.

Workflow systems

Existing research, diligence, committee, or monitoring systems that should remain part of the user workflow.

Implementation partners

System integrators and specialist partners that can help the institution operate and extend the environment over time.

Operating principle

Open-source first, clear boundaries, client control.

ZTA defines the operating boundaries required for governance and transferability while leaving room for the institution’s preferred tooling choices. Components remain replaceable where the substitution does not compromise control, auditability, security, or portability.

What is fixed

  • Read-only, entitlement-aware access patterns
  • Risk-tiered policy routing
  • Independent evaluation and human approval for consequential actions
  • Portable architecture and documentation

What can vary

  • Inference engine and serving stack
  • Selected approved open-weight models
  • Open-source software components
  • Adjacent workflow integrations
  • Infrastructure choices
  • Implementation and support arrangements