AI and automation¶
Zeta Test Management combines permission-filtered project data with centrally governed, provider-neutral AI services. The resulting workflows include project chat, quality analysis, test planning, knowledge guidelines, and safe domain actions from chat or MCP. Existing project roles and object visibility continue to apply.
Prerequisites and a safe starting sequence¶
- A server administrator creates at least one provider profile under Administration > AI administration and verifies its technical capabilities.
- The verified profile is assigned only to the required scenarios. Data zone, limits, and policy must match the intended use.
- External AI traffic is explicitly allowed when an external provider is used. Global AI is enabled only afterwards.
- A project administrator grants AI permissions only to the intended users, groups, or roles.
- Users review sources, evidence, limitations, and any proposed change before relying on an answer.
If an AI surface is missing, check more than sign-in. Global activation, scenario assignment, provider readiness, project approval, role, and object permission must all allow the operation.
Choose the workflow you need¶
- Configure and govern AI securely
- Use project chat and quality intelligence
- Work with AI test planning and knowledge guidelines
- Use safe actions, MCP, and Local Bridge
For the wider access-path comparison, see Product scope and interface coverage.
Start with a concrete task¶
| Your task | Suitable workflow |
|---|---|
| Understand progress or failure patterns | Ask project chat about the relevant plan, run and period; open its evidence |
| Check requirements for gaps | Select a requirement, run AI review and assess the findings |
| Improve or design test cases | Use AI case review or design in the library; review proposals before applying them |
| Generate cases from requirements | Select chapters, review AI candidates and commit approved candidates to a target unit |
| Draft a test plan | Use AI test planning with objectives, scope, risks and environments |
| Select tests for a change | Review AI test selection using release context, change intelligence and budget; hand the selection to run start |
| Make a supported project change | Prepare an offered domain action, review its preview and confirm |
For example, ask “Which failed tests in the current release share common causes?” or “Which requirements in the selected area show coverage gaps?”. Specify the relevant plan, run or requirement area and check that the answer uses that scope. Chat is not restricted to these examples.
Distinguish an answer, a draft and an applied change¶
An answer initially provides information. A saved AI test plan is an editable draft. Only an explicitly offered and confirmed domain action applies changes to project data. Check the target object and result afterwards.
The web interface supports more manual functions than the AI action catalogue. If chat or MCP does not offer an action, use the corresponding web area. Regular test management remains available without AI.
Responsibilities that remain with administrators and reviewers¶
AI does not bypass project security, approve its own provider, or turn a suggestion into a project change. Administrators remain responsible for provider and data-flow governance, while project roles remain responsible for access and final domain decisions.
- Keep provider secrets, MCP keys, and Local Bridge policies outside documentation and logs.
- Treat AI evidence and confidence as decision support, not as a replacement for acceptance criteria or professional judgement.
- Use the explicit preview and commit workflow for every supported write action.
- Disable external egress or global AI first when a data-flow or provider issue is unresolved.