Configure and govern AI securely

AI administration separates the model connection, scenario approval, and project-level permission. Zeta Test Management can use OpenAI, OpenAI-compatible endpoints including local models, OpenRouter, Azure OpenAI, Anthropic, and Gemini through the same gateway boundary.

Secrets remain server-side, are write-only in the UI, and are not returned in clear text. A successful connection check does not automatically enable external traffic or user-facing AI features.

Review AI results and control costs

Start project chat with a specific question. Review its sources and the current project state before relying on an answer. AI output may be incomplete or incorrect. Inspect proposed changes in the preview and apply them only after an explicit confirmation.

AI providers may charge their own fees; these do not change Zeta Test license prices. A chat may trigger multiple model calls. Set an additional provider spending limit where supported. Application usage figures do not replace the provider's invoice.

The public demo has separate AI limits. Visitors may share a demo account and its allowance. When that allowance is exhausted, AI remains unavailable until the next approval or budget period; this does not indicate a problem with your own installation.

Configure in this order

1. Open Administration > AI administration.
2. Use quick setup or create a provider profile with provider, data zone, endpoint, model, and API mode. Store the secret afterwards.
3. Run the capability probe and resolve reported incompatibilities.
4. Assign the verified profile only to the scenarios that should use it.
5. Configure governance, limits, and—where required—permission for external AI traffic.
6. Enable AI globally, then grant the project-specific AI permissions.

This sequence avoids exposing a feature without a ready profile or without a reviewed data-flow decision.

Provider profiles and scenario assignments

A provider profile represents one concrete technical model connection. Multiple profiles can separate data zones, models, or operational purposes. Scenario assignments select which verified profile may serve project chat, reviews, test planning, or another approved workflow.

• For local OpenAI-compatible services, use only endpoints approved and operated by your organisation.
• Never place API secrets in manuals, screenshots, browser consoles, or support logs.
• Repeat the capability probe after changing the endpoint, model, or API mode.
• Remove obsolete assignments and credentials.

Governance, health, and operations

AI administration exposes health, audit, usage, and cost information together with evaluation, chat-memory, and global knowledge-guideline controls. Project content is not treated as ordinary diagnostic log data. Use technical telemetry without copying confidential content into separate operational notes.

Review provider readiness, error rates, limits, data zone, and scenario assignments regularly. If the data flow is unclear, disable external egress or global AI before troubleshooting further. A healthy technical connection does not replace privacy, contractual, or permission review.