Whisper AI Assistant
Whisper is the in-platform AI assistant. It continuously surfaces compliance suggestions for the active version, and answers questions and proposes changes through an interactive chat.
Whisper is available from v3.2, but it is not enabled by default. Access is granted on request: contact Mia-Care (or your account manager) to enable Whisper for your organization. After access is granted, an administrator must turn on AI features and configure a chat provider before Whisper appears. See AI Configuration.
Opening Whisper
Whisper is available on every project-scoped page. Click the sparkles icon in the top navigation bar to open it. It opens as a panel on the right-hand side of the screen. You can expand it to full page or restore it to the panel at any time, and close it when you are done.
A context strip at the top of the panel shows the project and version Whisper is working against. Its suggestions and answers are scoped to that version.
The sparkles icon carries a badge showing the number of open compliance findings for the active version.
Compliance Suggestions
In the Suggestions view, Whisper lists AI-detected compliance findings for the active version, sorted by priority (High → Medium → Low). Each finding is a card with a title and a priority label.
Clicking a finding opens a detail drawer containing:
- Priority and Category
- Normative Reference: the regulatory clause the finding relates to
- Description and Reasoning: what was detected and why
- Remediation Steps: suggested actions to resolve it
- Confidence: how confident the analysis is in the finding
Running an analysis
Click Analyze now to run a fresh compliance analysis of the active version. The list shows a re-analyzing indicator while the job runs, then a confirmation once results are updated. A freshness label ("Updated N minutes ago") tells you when the analysis last ran. The list also refreshes when you return to the tab.
Triaging findings
Handle each finding from its card:
- Dismiss: mark a finding as not applicable. You can add an optional reason.
- Resolve: mark a finding as addressed.
After you dismiss or resolve a finding, an undo prompt lets you reopen it. By default the list hides handled findings and the badge counts only open findings. Use the Show dismissed/resolved toggle to see the full history.
Findings automatically reconcile when your requirements change: unchanged findings are preserved, findings whose underlying item changed are reopened, and findings that no longer apply are superseded.
Chat
Switch to Chat to ask Whisper questions in natural language. Answers stream in token by token, and conversations are saved.
- Conversations: a sidebar lists your past conversations; start a fresh one with New chat. Conversations are private to you within your organization.
- Composer: type your question in the Ask Whisper… box and send it.
- References: answers can cite the normative references they are based on, and every assistant message is marked as AI-generated.
Confirmable actions
Whisper can also propose changes to your project, for example creating or updating a requirement, or removing a risk link. It then presents a Proposed action card with Confirm and Decline buttons.
Whisper never changes your data on its own. A proposed action is applied only when you confirm it. Confirmed and declined actions are attributed to you ("AI-assisted · confirmed by you") and recorded in the project audit log.
From a finding to a chat
Start a chat directly from a compliance finding. Whisper opens a conversation seeded with the context of that finding, so it can suggest concrete remediation steps.
Choosing a model
If your administrator has made more than one chat model available, use the model selector in the chat to pick which model answers a given message. You can also refresh the list of available models from the configured provider. If a selected model is unavailable, Whisper falls back to the default model rather than failing.
When Whisper is unavailable
Whisper reports a message when it cannot run, for example:
- AI features are disabled: an administrator has not enabled AI for your organization.
- No active chat provider configured: AI is enabled but no chat provider/model has been assigned. See AI Configuration.