How does Velory use AI?

Summary and evidence

Velory AI is designed to give you two kinds of output:

• Summary: short text in chat that explains what the assistant thinks the answer is.
• Evidence: structured artifacts on the canvas, such as tables, charts, filters, and proposals.

The summary is useful for orientation. It may still be incomplete, mistaken, or based on a misunderstood request. Do not treat a sentence in chat as evidence on its own.

The evidence is what you should inspect. If the assistant understood your intent correctly, the canvas should show the facts behind the answer. Velory records remain the system of record, but the canvas is where you can see what the assistant used. If the prompt was ambiguous, the canvas also makes that easier to spot: check the records, dates, filters, and totals before you rely on the result.

What it is good at

• Finding records across hardware, software, employees, workflows, orders, access groups, tasks, renewals, and related lifecycle data.
• Turning questions into live tables and charts that you can filter, sort, search, and drill into.
• Answering follow-up questions from the current canvas state.
• Drafting workflows and supported action proposals.
• Reducing manual lookup when the result can be checked before use.

What it is not good at

• Being the source of truth by itself.
• Guessing the right scope when names, dates, teams, or filters are ambiguous.
• Making forecasts, budgets, projections, or strategic decisions.
• Accessing data beyond your permissions.
• Replacing human review for sensitive actions such as offboarding, access changes, renewals, or security work.

Changes and approvals

Velory AI does not make confirmed changes just because it says it will. Supported changes are staged as proposal cards on the canvas. You review the affected records and fields, then decide whether to confirm.

The chat text is not the approval. The proposal card is the thing to review.

After confirmation, the product applies the change through the normal Velory permission checks and audit trail.

Your data

Velory AI uses your existing Velory permissions. The assistant is given only the Velory data your account is allowed to access, and company data remains separated by company account.

Your prompt and the relevant slices of your company's data are processed to generate responses and artifacts. We do not use your data to train models.

If you use inline feedback on an AI answer, we snapshot the rated assistant message and the preceding prompt so we can review what was asked, what was answered, and what evidence should have been used without collecting the full conversation.

Avoid adding secrets, credentials, or unrelated personal data to prompts.

What to verify before acting

For read-only answers, inspect the canvas artifact before you act on the result. Check the rows, filters, dates, names, counts, and totals.

For proposed changes, review every record and field on the proposal card before confirming.

For MCP or external-agent output, compare the summary with the evidence rows. A good answer should be traceable back to the data it cites.

Reporting issues

If Velory AI gives an answer that is wrong, unsupported by the evidence, or based on the wrong intent, use inline feedback on that answer when it is available. Tell us what was wrong, what you expected, and which evidence looked missing or incorrect.

For anything urgent or sensitive, contact your Velory support channel.

Note: Velory AI is in early preview. Features may change. AI-generated answers should always be verified before acting on them.