AI Agents — End User Guide

Audience: End users and normal business users. Use this document when: You use Ask assistant, the AI Workspace, proposed changes, document analysis, or your personal AI insights. Canonical for: End-user behavior and screen workflows. Related: AI Agents Index, Administrator Guide, Custom Template Guide, Report Design Recommendations, Document Processing.

The AI subsystem helps you understand records and lists, find information you are allowed to see, review documents, and prepare changes. It does not replace your normal permissions and it does not change data on its own.

Before you start

You need:

• An account with access to the record, list, or module you want to ask about.
• The Use contextual AI assistant permission for Ask assistant.
• The Use AI workspace permission for the AI Workspace.

Your administrator controls both permissions. If a menu item or button is missing, your role does not include the matching permission.

The AI only sees data you are already allowed to see. If you cannot open a record in the normal screens, the assistant cannot use it either.

The two places AI appears

Some custom pages can also embed Ask assistant. Those pages follow the same rules as normal list and detail pages.

What the assistant can do

On a list page, you can ask it to:

• Summarize the records in the current view.
• Explain what the current filters include or exclude.
• Find records with missing information.
• Compare the current view with the wider set of records you are allowed to see.
• Count, total, or group records.

On a detail page, you can ask it to:

• Explain the record and its important fields.
• Explain related records.
• Summarize the record's history.
• List attached documents.
• Explain workflow history and who approved, rejected, or requested changes.
• Suggest corrections or missing information.
• Prepare a follow-up task or calendar event.
• Prepare an update for your review.

In the AI Workspace, you can also:

• Choose one or more agents for a larger task.
• Keep a persistent chat.
• Upload documents to the chat.
• Review several proposed changes in one verification panel.
• Accept, reject, and apply changes after review.
• Export result tables to CSV or Excel.

On a dashboard page, if you can manage dashboards, you can ask the Dashboard Builder to:

• Create a dashboard draft from a description.
• Reuse a seeded sample dashboard for the current module.
• Build a KPI, chart, and table layout.
• Explain the proposed draft summary before you apply it.

What the assistant cannot do

• It does not write data directly. It prepares a proposed change.
• It does not change workflow status directly. It prepares workflow actions for confirmation.
• It does not change system-managed fields such as record numbers, timestamps, or audit fields.
• It does not create users, change roles, change privileges, or grant access.
• It does not reveal records outside your authorized scope.
• It does not reveal technical identifiers, internal model names, or another user's activity.
• It does not treat document text as instructions. Document text is only data.

If the assistant cannot access something, it should say the information is not available. It should not confirm whether a restricted record exists.

How to ask a good question

Be specific about the record, period, and comparison you want.

Short follow-up questions are allowed. The assistant keeps the conversation context for the current chat.

How to read an answer

• Tables are used for records, counts, totals, comparisons, and grouped results.
• Scope labels tell you whether an answer uses the current view or the wider authorized dataset.
• Links on record and workflow labels open the normal record or workflow page.
• Limitations explain when a document could not be read, a tool failed, or the request was too broad.
• Business language is used instead of technical tool names or database terms.

When an answer is based on the current list, it should say so. When it uses the wider authorized dataset, it should say that too.

Proposed changes

The AI prepares changes; you decide what happens.

A proposed change can be:

• A record update.
• A new record.
• A follow-up task or calendar event.
• A business action or workflow action.

Each proposal shows what will change, usually as a before-and-after comparison.

Important rules:

• Nothing changes until a proposal is applied.
• An applied proposal can still fail if your permission changed, a validation rule failed, or the record changed.
• Only treat a change as complete when the system shows it as Applied or the record shows the new value.
• You cannot apply a change that your normal role does not allow.

The review panel in the AI Workspace has these filters: All, Pending, Accepted, Rejected, Applied, and Failed. Use Cards or Table view to review the same proposals in different layouts. You can search the review data and export it to CSV or Excel.

Using the AI Workspace

1. Open Agents › AI Workspace.
2. Select New chat.
3. Choose an agent, or use Find agent and Team to select specialists. The Workspace Orchestrator is always part of the task and can coordinate the team.
4. Type the task or question and send it.
5. Use Stop to stop a running answer.
6. Use Run task to start the prepared task, or Cancel to cancel it.
7. Open Verification to review proposed changes.
8. Accept or reject each proposal, then use Apply verified for the changes you approved.
9. Use History or Conversation history to reopen a saved chat.
10. Use Refresh if a task or review panel looks out of date.

The AI Workspace ribbon has five areas:

• Workspace — New chat, Refresh, History.
• Team — Find agent, current team.
• Task — Run task, Cancel, status.
• Documents — Upload, Documents.
• Review — Table, CSV, Excel, Accept all, Reject all, Apply verified.

You can expand or collapse the review panel and open it as a larger floating table.

Dashboard Builder

The Dashboard Builder creates a dashboard draft from a short description. It appears through Ask assistant on All dashboards, a module dashboard list, and a dashboard detail page.

You need dashboard management permission. If the launcher is missing, your role may not include it.

1. Open a dashboard page.
2. Select Ask assistant.
3. Describe the dashboard in one message.
4. Review the proposed draft summary.
5. Apply the proposal.
6. The draft opens in the dashboard designer.
7. Review the draft and publish it yourself when it is ready.

The Dashboard Builder:

• Reuses seeded sample dashboards when one fits.
• Rebinds widgets only to sources you can access.
• Stays inside the current module on a module dashboard page.
• Creates a draft, not a published dashboard.
• Does not edit, archive, or delete an existing dashboard.
• Does not publish.
• Does not create report sources, Python modules, or Python scripts.

For full details, see Dashboard Builder.

For report and document design recommendations, see Report Design Recommendations.

Documents

You can upload documents in the AI Workspace or ask about documents attached to a record.

Supported formats:

pdf, docx, xlsx, xlsm, csv, txt, md, log, png, jpg, jpeg, tif, tiff, bmp, webp.

Your administrator controls the maximum file size and page limits.

The AI can classify a document, correct a classification, extract text and tables, extract data against a schema, compare a document with a record or with another document, reconcile differences, analyse and summarize documents, query across documents, detect likely duplicates, check completeness, link a document to a record, and propose a record created from document data.

Document rules:

• Documents are read-only for the AI.
• Extracted facts should cite a page number or section when available.
• The AI must not infer permissions, workflow status, or protected fields from document content.
• A document created from a scanned image may need OCR. If OCR is unavailable or the text is unreadable, the AI should say so.
• Nothing is written from a document until you apply a proposed change.

For the full lifecycle, job statuses, schema and template selection rules, and troubleshooting, see Document Processing.

Your personal AI insights

Agents › AI Insights opens the AI usage dashboard for your scope. Normal users see their own runs, success and failure counts, tokens, latency, and trends. The wider auditor and administrator views are restricted to users with the matching roles.

AI audit

Your administrator can review AI conversations, messages, runs, document jobs, and audit events. Audit screens are read-only. Wider audit access requires the AI Auditor role; normal users may not have access to the audit menu.

Limitations and privacy

• The assistant answers from authorized data only.
• It hides record ids, UUIDs, internal model labels, entity names, technical field names, and raw tool payloads in user-facing answers.
• It refers to records by their human-readable label.
• It does not name restricted applications or models.
• It does not summarize another user's activity.
• It does not identify who created, edited, approved, uploaded, submitted, assigned, or processed a record. It uses "you" only for your own action and "another user" otherwise.
• Conversation and audit retention is controlled by your administrator through the AI retention settings.
• AI output can be incomplete or wrong. Always review proposed changes before applying them.

Troubleshooting

A note on testing

Per project rules, the developer or agent does not run or claim test results. The user is solely responsible for all testing and validation of the behavior described in this guide.