Spot likely personal data
The dataset scanner flags patterns such as email addresses, card numbers and New Zealand phone, IRD, NHI and bank-account details. Findings help your team review and classify data before publishing it.
Nexus overview ↗AI with human review
Human-governed AI for quality engineering.
Analyse a requirement when useful or generate test drafts directly. Review every finding and draft before it becomes part of the Jira record.
Get started →- Requirement context
- AI drafts
- Human review
- Accepted tests
- Usage visibility
Human-reviewed test drafts
Connected intelligence
Choose the model.
Keep the workflow.
Nexus Pro connects to your own AI account. Choose a supported provider and model centrally in Nexus Setup.
NEXUS PRO / CONNECTED INTELLIGENCEYour provider. Your account.
THE WORKFLOW LIVES IN JIRANexus
Review requirementsDesign testsApprove the outcome
One active provider & model
- AnthropicClaudeSonnet · Opus
- OpenAIGPTGPT models
- GoogleGeminiPro · Flash
- Moonshot AIKimiK-series models
- xAIGrokGrok models
One connected workflow.Your administrator chooses the provider and model in Nexus Setup. Model availability depends on your provider and account. Families shown are examples.
Your account, your usage. Model usage is billed by your provider. Nexus reports token usage and keeps test drafts under human review.
Rovo has its own connection. The Nexus analyst runs through Atlassian Rovo and your Jira permissions. Explore Rovo support →
PII & sensitive-data protection
Control the data.
Choose what you share.
Nexus combines personal-data scanning, sensitive-value masking and rules for which dataset rows can become AI examples. Review requirements and test-case content before sharing it with your chosen AI provider.
Explore test data controls →EXISTING DATASET CONTROLSWhich rows can become AI examples?
Synthetic rowsOnly with your opt-inUp to five example rows
Masked rowsExcludedNo row values sent as examples
Sensitive rowsExcludedNo row values sent as examples
Only synthetic rows can be included as examples, and only when you choose to include them.
Control sensitive values
Sensitive datasets show masked values by default. Owner and owner-group controls govern explicit reveals, with reveal activity recorded. A masked dataset means identifiers have already been replaced; choosing that classification does not anonymise the data for you.
Limit AI example sharing
For AI test-data suggestions, existing rows are optional. Only rows classified as synthetic can be used as examples, with a maximum of five. Masked and sensitive rows are excluded even when examples are requested.
01 / Optional requirement review
Find the questions before they become defects.
When analysis is useful, review requirements for ambiguity, gaps, contradictions and weak acceptance criteria. Findings point back to the requirement so your team can investigate the evidence and decide what needs clarification.
Analysis is optional; it is not a gate before test generation. Keep useful findings, correct assumptions and dismiss suggestions that do not apply.
02 / Test case creation
Turn requirement intent into reviewable tests.
Generate traced test-case drafts directly from a requirement. Approved findings can inform the draft when they are available, and your team reviews every proposal before deciding which cases belong in the test library.
Review the coverage
Inspect the suggested scenarios, requirement links, actions and expected results. Check that the tests prove the behaviour your team intended.
Edit, accept or dismiss
Correct the draft, accept useful tests and dismiss irrelevant suggestions. Generating a draft is separate from accepting it as a Jira test case.
Keep a reusable library
Accepted cases become part of your Jira test library, ready for planning and execution. An AI-generated case is a design, not evidence that the behaviour passed.
03 / Model-based testing
Keep the model under your control.
Use state machines and decision tables to express the behaviour your tests need to cover. Review the model, refine its rules and generate test cases from the saved structure.
- 01
Model the behaviour
Describe states and transitions, or map conditions to expected actions in a decision table.
- 02
Review the rules
Check the paths, conditions and outcomes before saving the model.
- 03
Generate from the model
Use the saved model to generate test cases. The model rules drive repeatable generation.
Model-based generation is deterministic and works without an AI provider account.
Explore the connected workflow04 / Test data assistance
Turn test inputs into reusable data.
Use AI to help prepare the data your test cases need, with a review step before applying suggestions.
Suggest data. Review the detail.
Propose data columns and scenario rows, or find data written into steps and suggest reusable placeholders. Review, edit or reject suggestions before applying them to a dataset or step.
Masked or sensitive dataset rows are not sent as examples.
Explore test data →05 / Rovo support
Ask the Nexus analyst in Rovo.
Use Atlassian Rovo as another entry point to Nexus. The agent works within your Jira permissions, so answers reflect the work you can access.
Rovo must be available and enabled for your Atlassian site. Its availability is separate from the AI provider account connected to Nexus Pro.
Ask about your testing evidence
- “What is the coverage for this requirement?”
- “How is this test cycle progressing?”
- “What needs attention before this release?”
- “Give me an overview of this project.”
06 / Token and usage management
See the usage.
Choose the controls.
Connect your own provider account and monitor recorded AI usage in tokens. Nexus brings model configuration and recorded usage together for administrators to review.
Input, output and run history
View recorded input and output tokens, run counts and monthly history. Project breakdowns help administrators understand where usage comes from.
Warnings before surprises
Configure a site-wide monthly token threshold and sudden-usage-spike alerts. Warnings appear in Nexus for administrators to review and dismiss.
Site-wide AI control
Manage the provider and model centrally. Turn AI off for the site when it should not be used.
Choose the model's thinking level where supported, balancing reasoning depth, response time and token use.
Token alerts are warnings, not spending caps.
They do not stop or slow AI runs. Nexus reports tokens, not a currency bill, and does not impose its own AI usage cap. Your provider bills model usage separately from the Nexus subscription; manage provider-side budgets and limits in that account.
Your provider. Your decisions.
AI suggests. Your team decides.
Choose your provider
A Jira administrator connects Anthropic, OpenAI, Google Gemini, Moonshot Kimi or xAI Grok for the site. The provider key stays in Forge secret storage, not in the browser.
Core test management and deterministic MBT work without Pro AI features. AI output remains subject to human review and is not a release sign-off.