Optiak Docs
Compliance

Data Governance

Define how your organization's data is handled across AI applications.

Data Governance centralizes the policies that protect your organization's data across AI applications. Set your requirements in one place so applications use models and tools that meet them.

Required role

Organization Admin

Applies to

LLM requestsTool calls
Organization Admins set these policies once for the whole organization. They apply to every application, and App Admins can't change them or opt out.

Why use it

Regulations and contracts often say where personal data may be processed and forbid vendors from keeping it or using it for training. Writing those rules into every application is slow and easy to get wrong. With Data Governance, you state them once and the Gateway applies them when selecting eligible models and tools.

  • Enforced, not advised. A model or tool that doesn't meet the policy is never called, whatever the application asks for.
  • One switch for every application. New applications inherit the policy automatically.
  • Evidence on every request. Each request record shows the policy in force and the data residency of the model that served it, so you can show an auditor where each request went.

How it works

Set your organization's policies once. The Gateway checks models, tools and Data Integrations against the policies that apply to each service.

Data residency applies to models, tools and Data Integrations. Zero data retention and Block providers for training on your data apply to models. They do not enforce retention or training restrictions on external tools or Data Integrations.

Organization policies

One set of rules for every application

Models

Must meet every active model policy

  • Data residency
  • Zero data retention
  • No provider training

Tools & Data Integrations

Must meet the active residency policy

  • Data residency

Gateway decision

Compare provider labels with the applicable rules

Eligible

Available to the application

Excluded

Cannot be used under the policy

The Gateway applies the active policies when selecting a model or tool for a request. An Organization Admin can enable a model that does not meet those policies, but the Gateway excludes it from selection. If no enabled model meets the policies, the LLM request fails.

If a request explicitly names a model that does not meet the active policies, the Gateway rejects it with 403 Forbidden and lists the policy's requirements. For example: Model '<model>' is not available under your organization's governance policy, which requires EU data residency and zero data retention.

Review these labels under Company setup. Organization Admins manage declarations for the services they connect.

ServiceYour action
  • Built-in models
  • Data Integrations
Review the labels. Optiak sets them, and they cannot be edited.
  • Preintegrated external MCP servers
Review the suggested residency. Update it if your provider agreement or deployment differs.
  • Custom providers
  • Custom MCP servers
Declare the labels based on your confirmed arrangement.

For preintegrated external MCP servers, Optiak suggests residency defaults based on each provider's standard service. Your account, deployment or agreement with the provider may have different residency terms. Review the suggested default before using it, and set the label to match your confirmed arrangement.

Labels you declare are your organization's statement about that provider. Choose Unknown if you can't confirm where a server processes data. With No preference, an unknown residency does not exclude the server. With a region policy such as EU / EEA, it is excluded from the application's available tools and cannot be used while that policy is active.

Each LLM request's step-by-step trace, in the application's Requests tab, records the governance policy applied when the model was chosen and the data residency of the model that answered. Changes to the policies are recorded in Audit Logs.

Policies

Data residency

Data residency keeps processing in the region you select. It applies to LLM requests and tool calls: the models that answer, the tools applications can call and the data integrations that feed their prompts.

SettingEffect
No preferenceAny enabled model, tool or data integration can be used, wherever it runs.
A region, such as EU / EEAOnly models, tools and data integrations labeled with that exact region can be used. Those labeled Global or with no known residency are excluded.

Tools and data integrations that don't match are hidden from applications: App Admins can't enable them, and ones already enabled stop being offered.

No training on your data

Block providers for training on your data excludes every model whose provider may use the requests it receives to train its models. It applies to LLM requests. In the model catalog, qualifying models carry the No training badge.

Zero data retention

Zero data retention excludes every model whose provider may store requests and responses after answering. It applies to LLM requests, and is the strictest choice when you can't allow copies of prompts or answers to remain with a provider.

Set it up

Set these policies once for the organization. They apply to every application, with no per-application configuration.

In the console, open Company setup, then Governance. The three policies are under Data & privacy, and each change is saved as soon as you make it.

  1. Check which models, tools and data integrations meet the policy you want, using the labels described in How it works.
  2. Make sure every application still has at least one enabled model that qualifies. Each application chooses its models in its Models settings; see Models.
  3. Choose the Data residency region and turn on Zero data retention or Block providers for training on your data as needed.
Tightening a policy can leave an application with no eligible model, and its LLM requests then fail. Check each application's enabled models before changing a policy.

Limits and behavior

  • Defaults. Data residency is No preference. Zero data retention and Block providers for training on your data are off until you enable them.
  • smart or no eligible model left: Smart Router chooses only among models that meet the policy. If none of the application's models qualify, the request fails.
  • Policies combine. With several policies on, a model must meet all of them.
  • Changes apply from the next request after they're saved.
Copyright © 2026