Ask and enrich with the AI assistant

The assistant can help you understand the knowledge base and prepare changes. A written answer, a proposed action and a saved change are different results: check which one you received.

Availability depends on your organization's plan, AI settings and your own access. Ask an administrator to check AI configuration if the assistant is unavailable.

Ask a contextual question

1. Open the organization and choose Real data or the intended scenario.
2. Open AI Assistant from the relevant screen.
3. Start a New conversation or select an existing conversation in the panel.
4. Choose an available model if you want to change the current selection.
5. Write your question, naming the relevant assets, type and expected result. Use @ to select a reference when appropriate.
6. Send the message and read the answer, referenced elements and any proposed actions.
7. Open the relevant asset pages to check the information before making a decision.

For example: “List the applications linked to this capability and show their allocation percentages. Do not change any data.” Selecting the exact asset is more precise than using a name that several assets might share.

Ask for counts, distinct-value counts, sums or averages, with a breakdown by one or two dimensions. For example: “Total annual application costs by hosting mode, then by criticality.” These requests only read data. Multi-valued groups can overlap, so their results do not always add up to the overall total.

Ask what is waiting for you, for example: “What do I have to do in Boldo?” The assistant reads the To do and My suggestions tabs of your Home: suggestions to review, assets to attest or complete when repository quality is active (attestation in real data only), and the suggestions you sent, pending or resolved in the last 30 days. It answers with links to the matching screens and does not change any data. Its Reasoning block shows Read the to-do list.

You can also ask from search: type your question, then choose Ask the assistant. It opens a new conversation in the panel, with the current page as context.

Use Open in full page for the conversation history and AI projects. Close closes the side panel; it does not delete the conversation. You can reopen the panel from AI Assistant.

Follow or retry a response

While the assistant works, its activity line shows the current step, elapsed time and number of tool actions. Click it to expand the details. The line stays collapsed by default; an activity count does not mean that changes were saved.

If a provider error or timeout interrupts the latest response, use Retry when offered. This action also remains available after reopening the conversation. Check any partial result and already applied proposals before retrying.

A message explaining that the response reached its maximum length means the answer was cut short. Ask for a shorter answer or split the request into smaller tasks.

Choose a suitable model

The model selector lists the options available to your organization. Recommended identifies the configured recommendation; it is not a guarantee that every answer is correct.

For a custom model, the selector also shows the service host configured by your organization. This helps identify which service will receive the request.

Models can differ in their support for images and actions. Use a model marked as supporting images when attaching a picture. Switching to an incompatible model removes an image from the current draft. Selecting a compatible model again does not restore that removed attachment.

For documents and limits, see attachments and project documents. The models shown can change with the organization's configuration, so use the current selector instead of relying on a fixed list of model names.

A model or endpoint that cannot use Boldo's tools may answer in text without performing the requested operation. Check the returned action and result rather than assuming that a confident answer changed the knowledge base.

Prepare an asset form

The AI input inside an asset form works on that form's draft. It can prepare the name, supported properties and relationship additions or removals. The form still needs to be validated to save those changes.

It also works when creating or correcting a creation suggestion. In Suggestion mode, it prepares the values you can suggest. Send the suggestion submits the draft for review; Save on an existing suggestion updates that suggestion. Neither action creates the inventory asset.

1. Open Create asset or Edit on an existing asset.
2. In Ask AI to create, fill, or modify…, describe the changes you want.
3. If you need a specific related asset, enter @ and select the exact reference.
4. Send the request and wait for the prepared result.
5. Inspect the actual fields, their previous values and the Relationships tab. Adjust anything incomplete or incorrect.
6. Use Undo AI changes if you want to return to the form state before the last AI turn that changed fields.
7. Review the complete form, then select Validate, Send the suggestion or Save, according to its mode. Cancel abandons the draft.

For example: “Set the supplier to Example supplier and link this application to the selected capability. Keep the other properties unchanged.” Afterward, check both the supplier field and the link.

Do not assume every part of the request was fulfilled. The form assistant does not populate unique-key properties through this workflow. Fill those identifiers manually when needed. It also does not offer a general way to clear properties by requesting null values.

This input adds or removes links without populating their relationship properties. Open its Variables and complete them before validating. For a precise change to an existing parallel relationship, review a relationship proposal or edit the intended instance directly.

Undo, dismiss and stop

Undo AI changes restores the snapshot taken before the last turn that modified the form. It can also remove manual edits made after that snapshot. A later text-only answer does not replace the previous restore point.

Dismiss removes the answer and access to that undo action. It keeps the prepared field values and does not save them. This differs from Cancel on the parent form, which abandons the draft.

Stop generating interrupts the current response. It does not undo a change already saved elsewhere or guarantee that no partial response remains. Inspect the form and conversation before continuing.

Keep the right context

A conversation belongs to the real data or scenario context in which it was created. Changing the application's current context does not transfer that conversation.

When you open it from another context, the composer and proposed actions can be locked. Use the offered action to return to real data or open its scenario. If the scenario is archived, applied, abandoned or unavailable, the conversation can remain read-only. Reactivating an archived scenario can let you resume it; no proposal is applied automatically.

A change created in a scenario remains there until you use the scenario application workflow. Applying an AI proposal inside a scenario is not the same as applying that scenario to real data.

The assistant's available actions also depend on your permissions. When it can only prepare a suggestion, sending that suggestion does not apply the final business change. Follow the proposal review procedure and inspect the recorded result.