Asking Questions¶

Availability

Who can do this: members with the Chat capability in the active Workspace.

Starting a conversation¶

In the Explorer sidebar, select New task.
In the sentence Ask <analyst> a question, open the analyst picker and choose an AI Analyst. Before you pick one, the picker reads Choose an analyst.
Type your question in the message box, or choose one of the questions under Or try a suggested question:. Up to three suggestions are shown. When there are more, select View more to open Suggested questions, with one tab per category.
Press Enter to send. Press Shift+Enter to add a new line instead.

Picking a suggestion only loads it into the message box. Nothing is sent until you send it, so you can edit it first.

You don't need to phrase questions in any particular way. Write as you would to a colleague. For tips, see Tips for Asking Questions.

To start from one of your team's Saved Prompts, type / at the start of an empty message and pick a prompt.

Choosing an AI Analyst¶

The picker lists only the AI Analysts you're allowed to use. Members with only the Chat capability see deployed analysts shared with everyone in the Workspace or with them specifically. Members with the Edit capability also see draft analysts, marked Not deployed.
Select the star icon (Star) next to an analyst in the picker to star it. Select it again (Unstar) to remove the star.
Explorer remembers the last analyst you picked in this browser and selects it for your next new conversation. If you can no longer use that analyst, it's cleared.
Data team members can open an analyst directly from Studio: in AI Analysts, open the analyst's row menu and select Open in Explorer. Explorer opens a new conversation with that analyst selected, as long as you're allowed to use it.

Once you send the first message, the conversation is tied to that analyst. The message box shows the analyst's avatar and name instead of the picker, and you can't switch analysts in that conversation. To ask a different analyst, select New task.

If you're no longer allowed to use the analyst when you send, the conversation shows: "You can't start a conversation with this analyst — it's no longer available to you."

The conversation link¶

Sending the first message gives the conversation its own address. Reloading the page, using the browser's back button or opening a copied link brings you back to the same conversation. Other people can open the link only if you share the conversation with them.

Watching the AI Analyst work¶

While the AI Analyst works, a status line shows what it's doing, such as Thinking…, Looking up your data models…, Writing a query… or Running the query….

Its steps then appear in a group labeled Working…. When it finishes, the group folds and reads, for example, Worked for 12 seconds · 4 steps. Expand it to see each step, such as Run SemQL query, Create chart or Present.

If the AI Analyst had to make assumptions about scope, such as a time window or a filter, an Assumptions block lists them under the answer. Key figures in an answer are highlighted so you can spot them quickly.

For more about steps, assumptions and how to check an answer, see How answers are built.

Conversational follow-ups¶

Explorer keeps the context of the whole conversation, so you can reply with a follow-up that builds on the previous answer, or explore a related angle without repeating yourself. Start broad, then drill into the areas that matter.

Stopping and retrying¶

While the AI Analyst is working, the send button turns into a stop button (a square icon). Select it to stop the answer. The answer is marked Stopped.
If something fails, the conversation shows "Something went wrong running this question. Please try again." Send your question again, or rephrase it.
When the failure came from the language model, a Retry button appears. It sends the same question again. See When the AI Analyst can't answer.

Plan Mode¶

For complex or high-stakes questions, turn on Plan Mode with the Plan button in the bottom-left of the message box. With Plan Mode on, the AI Analyst shows a Suggested plan before running any analysis. You can approve it, edit it, ask for a different plan, or cancel.

Plan Mode is off by default. Your choice is remembered in this browser for new conversations. Each conversation keeps its own setting. Approving or cancelling a plan switches Plan Mode off again.

Saving results as Artifacts¶

Charts, tables and reports produced in a conversation can be saved as Artifacts: named outputs that live in your Explorer library.

Select Save on a chart or table card in the conversation to save it.
Once the conversation has a report, select Add to report on a chart or table card to add it to the end of that report.
Select the bookmark icon (Save artifact) at the top of an open report to save it.

Saved artifacts appear under Artifacts in the Explorer sidebar. See Artifacts for details.

Building reports¶

The AI Analyst can compile your analysis into a structured report, with headings, paragraphs, charts and tables, directly in the conversation. Reports appear in a panel next to the conversation and build as the AI Analyst writes.

Ask for a report whenever you want a structured write-up, for example: "Can you pull this together into a report?" The AI Analyst usually builds one when you ask for a review, briefing or multi-section summary.

See Reports for details.

Conversations are private by default: in Explorer, only you can open them. Members with the Edit capability can still review every conversation with the Workspace's AI Analysts in Studio Monitoring. See Sharing.

To share a conversation:

Open the conversation and select Share.
In Share conversation, choose Only me, Everyone in this organization or Specific people.
Select Save, then use Copy link to copy the conversation's link.

Only the conversation's owner sees Share. Copy link is available once you've saved an audience other than Only me. Everyone in this organization means every member of this conversation's Workspace, not the whole organization.

People you share with must have access to the conversation's Workspace. They see the conversation read-only: only the owner can send messages. See Sharing.

Managing your conversations¶

The Explorer sidebar lists your conversations:

Section	What it contains
Starred	Your own conversations that you starred.
Conversations	Your other conversations, most recently updated first.
Shared with me	Conversations other people shared with everyone in the Workspace or with you specifically. This section appears only when it has items.

To manage a conversation, hover over its row and open the ⋯ menu (Conversation actions):

Star / Remove star: stars are personal. You can also star a conversation shared with you; it stays under Shared with me.
Rename: change the conversation's Title, then select Save. Available only for your own conversations.
Delete: available only for your own conversations. Confirm with Delete, or select Keep it to cancel.

Deleting a conversation removes it, with its answers, from your sidebar, and everyone it was shared with loses access. You can't restore a deleted conversation from Explorer. Members with the Edit capability can still see a deleted conversation in Studio Monitoring.

Sidebar limits

The sidebar shows up to 50 of your most recently updated conversations, and separately up to 50 conversations shared with you. Starred is drawn from your 50 most recent conversations, so an older starred conversation can drop out of the sidebar.

Scheduled insight runs¶

Each run of a Scheduled Insight creates its own conversation, owned by the schedule's owner and titled with the schedule name and the run date. These runs appear under Conversations like any other conversation. When you open one, a strip at the top reads From scheduled insight followed by the schedule name. Select the name to open the schedule.

Read-only conversations¶

Some conversations can be read but not continued. A banner explains why:

Banner	When you see it
This conversation was shared with you in read-only mode. Only the owner can send messages.	Someone else owns the conversation and shared it with you.
This analyst is no longer available to you. You can read this conversation, but can't send new messages.	The analyst was undeployed or deleted, or you were removed from its audience.
This conversation was migrated from the previous product. You can read it, but you can't send new messages. Start a new conversation to keep asking.	The conversation was brought over from the previous product. The message box is hidden.

In a read-only conversation, the Plan button, plan actions, Retry and feedback controls are not available. To keep asking, select New task.

Usage and message limits¶

If your organization's plan has a message limit, a Usage button appears in the message box. Hover over it to see how many messages have been used and how many are left, and when the allowance renews. It isn't shown when your organization has no message limit.

When your organization reaches its limit, new questions are blocked with: "Your organization has reached its message limit — contact your admin to upgrade."

When the AI Analyst can't answer¶

These messages mean the language model behind the AI Analyst couldn't be reached. They depend on how your organization's inference is set up.

Message	What it means	What to do
Chat needs an LLM endpoint	No language model is connected for your organization. This usually happens only on a self-managed deployment where no LLM endpoint has been set up.	Ask your administrator to connect an LLM endpoint, then select Try again.
Your organization's own LLM endpoint didn't answer. An administrator can check it in Settings → Inference.	Your organization uses its own LLM endpoint, and it didn't respond. AI Analyst doesn't fall back to another model.	Select Retry. If it keeps failing, ask an administrator to check Settings → Inference.
Actian Inference couldn't answer this question. Please try again in a moment.	Actian Inference failed to answer.	Wait a moment, then select Retry.

Try again and Retry appear only in conversations you can reply to. The Test your analyst panel in Studio shows the same messages without these buttons; send the question again to retry. Steward in Studio shows similar messages; see Using Steward.

Multilingual support¶

Actian AI Analyst understands and responds in the language you write in, with no configuration required. Ask your question in French, Spanish, Japanese, Arabic or many other languages, and the AI Analyst replies in that same language.

Team members can each ask in their own language, even when using the same AI Analyst. The language of the interface itself (menus and buttons) is a separate setting in your Profile.

Investigating unexpected answers¶

If a result looks wrong:

Expand the step group under the answer to see what the AI Analyst did, and check the Assumptions block. See How answers are built.
Rate the answer or select Report a problem under it. See Rating answers and reporting problems.
Data team members with the Edit capability can open the AI Analyst in Studio, select Monitoring → Tasks, and pick the conversation to replay it and inspect each step. See Monitoring.