When the AI can't answer

When the AI refuses a request or stops, the message says why. Find its title below: what it means, who can fix it and how.

A request refused before the AI started uses no AI credits. A request can also end partway, because it took too long, failed, was stopped or was refused at a later step; the AI work it had done by then is charged (AI credits).

AI turned off or not for you

These are the workspace’s AI controls, which a company administrator sets under Administration → AI (AI controls). Each refuses the request before the AI started, and only AI is refused: dashboards, reports, queries and everything else keep working.

AI is turned off here

AI is turned off in this workspace, or AI for card summaries is turned off in this workspace (or another feature). An administrator turned AI off for the whole workspace, or for that AI feature. Administrators are told where to turn it back on; everyone else is asked to ask an administrator (Turn AI off for the workspace or for one feature).

Where AI is only one step of a feature, such as Explain this change or a profiling scan, the feature still runs without its AI step.

AI isn’t available to you here

Your role doesn’t include AI for pipelines (or another feature), or You’re not a member of this workspace, so its AI isn’t available to you. A company administrator chooses which roles use which AI features; ask an administrator if you need it (Choose who may use AI).

AI is paused for a moment

AI is paused for a moment: its settings couldn’t be checked. DataSquares couldn’t read the workspace’s AI controls just now, so it refused the request rather than run AI they might not allow. Try again shortly.

Credits, budgets and limits

AI credits used up

This workspace has used its AI credits. The workspace’s balance is at zero, so the AI can’t start a request, or the next step of one it had started. If the AI had already done part of the work, that part is charged.

• Administrators get Buy credits, which opens AI credits on Administration → Billing: buy a credit pack, turn on auto top-up so a pack is bought when the balance runs low, or let AI go below zero up to a limit, billed on the next invoice (AI credits). If the message says the company’s limit below zero is used up too, buy credits or wait for the next month.
• Viewers use their own credits first, then the credits the company bought, not the editors’ included credits, so a viewer can be refused while editors still have credits (You have used your own AI credits for this month, and the credits your company bought are used up too.). Their own come back next month; before then, ask a billing administrator. If an administrator turned off Viewers may use the credits your company bought once their own are used (Administration → AI), a viewer whose own credits are used up is told AI for viewers from the credits your company bought isn’t turned on, and has no more AI until next month.
• Everyone else is asked to tell a workspace admin.
• The monthly allowance also comes back on its own when it resets.

Applying a change you had already approved still works: it runs no AI (Applying it).

Monthly AI budget reached

You’ve used your monthly AI credit budget in this workspace, or This workspace has used its monthly AI credit budget. An administrator set a monthly budget of AI credits for you, for each person or for the whole workspace, and it is used up, so the request was refused before the AI started. AI pauses until the first of next month (UTC), which the message names. Administrators can raise or remove the budget, and AI runs again at once; everyone else is asked to ask an administrator (Set monthly AI credit budgets).

AI usage limit reached

A usage limit on AI was reached. The message says which:

• Your own hourly limit, which frees up as the hour rolls on; the message says about when.
• The workspace’s limit for today or this month, or the company’s for this month. It resets on its own at the start of the next one.
• The workspace’s limit for now, which resets on its own.

None of these is a setting in your workspace (the budgets an administrator sets are, above). The hourly, workspace and company limits are checked before the AI starts, so a request they refuse uses no credits. The limit for now can also stop a request partway: if the AI had already done part of the work, that part is charged.

AI usage service unavailable

DataSquares couldn’t check the workspace’s AI limits just now, so it paused the request rather than risk going past them. The AI didn’t run. Try again shortly.

Not offered here

In Chat options, a level above the highest your company allows reads Not offered here. Every plan, a trial included, has every effort; an administrator sets the highest one. Choose another effort, or ask an administrator (Effort).

This effort level isn’t available right now

This request needs an AI effort level that isn’t available to this workspace right now. The AI service DataSquares uses refused the effort this request needed. It is a setup problem on DataSquares’ side, not your plan or your credits, and the team is notified. There is no effort for you to change: where it can, DataSquares has already asked at a lower one. As the message says, Please try again later. Refused at its first step, the request used no credits; if the AI had already done part of the work, that part is charged.

The AI is unavailable

AI is temporarily unavailable

AI is temporarily unavailable right now. The AI service DataSquares uses isn’t taking requests right now, and the team is notified. Try again later; your question stays in the box. Refused at its first step, the request used no credits; if the AI had already done part of the work, that part is charged.

AI is catching up

Answers may take longer than usual. A question that cannot reach the AI right now waits and continues on its own. The AI service is answering more slowly than usual, or not at all for a moment. Keep asking: on a server that turns plans on, a question that can’t reach the AI waits for up to ten minutes and goes on by itself, its working card saying The AI is busy right now; it will continue on its own. The banner clears by itself once the service is back. Waiting uses no AI credits.

The AI could not be reached for a while

(The AI could not be reached for a while, so this reply stops here with what was found so far. Ask again to continue.) The question waited ten minutes and the AI still couldn’t be reached, so the reply keeps what was found, labelled Response incomplete. It also ends this way sooner when the AI keeps failing on the same step: after three tries in all, so that a step that fails each time is not tried again and again. Ask again once AI is catching up has cleared; the AI work it did is charged, each try for what it did, and the waiting is not.

AI service unavailable

The AI service didn’t answer, or stopped answering partway. Retry sends your message again. If the AI had already done part of the work, that part is charged.

Couldn’t reach the AI service

Your browser couldn’t reach DataSquares. Check your connection and retry; the request stays in the box. If an answer had finished, the chat restores it from the server (Practical notes).

Reconnecting…

On a server that turns on picking answers up again, an answer keeps going while your connection is down, and the working card says Reconnecting… until the chat is back in touch; then it shows what you missed and carries on. Nothing to do: if your connection stays down for long, the answer is in the conversation when you come back (When something goes wrong).

AI isn’t set up on this server

On a DataSquares server your organization runs itself, AI needs to be configured by the people who run it. Administrators see AI isn’t set up on this server; everyone else sees AI isn’t available in this workspace yet. Ask the server’s administrators.

Access and data

You don’t have access to complete this request

Your role doesn’t allow the action, or the data it needs. Choose data and an action your account allows, or ask a workspace admin for access (Users, roles & invites).

Not available for your account

Your data permissions don’t allow this analysis. Explain this change and Analyze show this when a row-level security policy on the card’s source applies to you; Summarize with AI refuses you in that case too (Card insights).

This request is based on out-of-date data

The dashboard, report or data the request refers to has changed since it was made. Review the latest version, then send the request again. On an approval card the same thing reads Not applied, with the reason: ask for a fresh preview (Applying it).

The selected data is no longer available

The data source or model the request named was deleted or can’t be found. Choose a current one and send the request again.

This suggestion is out of date

The suggestion you clicked no longer matches the answer or the page it was offered for. Pick one from the latest answer, or ask your question directly.

This request needs to be adjusted

The server couldn’t accept the request as it was sent. Review it, and for an AI action its scope and destination too (The selected AI action needs to be adjusted), then send it again.

Your session has expired

Sign in again to continue.

Long or complex requests

The response took too long

Please try a simpler request. An answer in AI chat stops after two minutes. Split the request into smaller ones. The AI work done until it stopped is charged.

On a server that turns plans on, each step has a time limit of its own instead, and a whole answer stops after 15 minutes. A query that runs too long is stopped and the AI works around it. When the AI itself takes too long once something was found, the answer is written from that and ends (A step took too long, so this answer uses what was found so far. Ask me to continue, or narrow the question.) If writing that answer takes too long as well, what it had written stays and the reply ends (A step took too long, so this answer was cut off before it was finished. Ask me to continue, or narrow the question.); when it had not begun the answer, the reply ends (A step took too long, and no answer was written from what was found. Ask me to continue, or narrow the question.). Either way, what was found stays in the steps. When nothing was found yet (writing the plan does not count), the request ends with A step took too long, and nothing was found yet to answer from. Ask again, or narrow the question. Ask the AI to continue, or narrow the question. The AI work it did is charged, like any other.

This reply was interrupted and could not be finished

(This reply was interrupted and could not be finished, so it stops here with what was found so far. Ask again to continue.) The service restarted in the middle of the answer more often than the answer could pick up again, so it ends with what was found, labelled Response incomplete. Ask again; the AI work done before each interruption is charged once.

Rarely, the reply ends with the message This reply was interrupted and could not be finished. Ask again to continue. instead: the service restarted and could not pick the answer up again at all (what it had kept could not be read back, or an update came before anything was kept). Ask again; the AI work it did is charged once, like any other.

You asked for the answer now

(You asked for the answer now, so this answer uses what was found so far.) You pressed Answer now while the AI worked: it stopped between steps, stopped the queries still running, and answered from what it had found, saying what it did not get to (While it works). During a build, what is ready is shown for approval, and you can ask the AI to continue with the rest. When nothing had been found yet, the note says so: ask again, or narrow the question. The AI work it did is charged, like any other.

This question needed more steps than one answer allows

One answer can take only so many steps. When a request needs more, the AI stops looking and answers from what it found so far, and the answer ends with this note and says what it could not finish. Ask the AI to continue, or narrow the question. The answer is charged for the AI work it used, like any other (What the AI can and cannot do).

When the AI ran out of steps before it could write anything, the reply says (This question needed more steps than one answer allows, and no answer was written from what was found. Ask me to continue, or narrow the question.) Ask for one part at a time. The AI work it did is charged, like any other.

Rarely (an older server, or a request allowed no steps at all), you see This needed more steps than one answer allows instead, with Narrow the question, or ask for one part at a time. Your request stays in the box: narrow it, or ask for one part at a time.

This conversation is too long for the AI to read in one go

Start a new conversation, or shorten your message.

The AI couldn’t process this request

The AI service turned the request down as it was written. Try rephrasing it, or start a new conversation.

Approvals

Approved, not confirmed as applied yet

The connection dropped after you approved a change. Try again sends the same approval, and a change is applied only once: if the first one went through, you see Applied successfully (Applying it).

Ask AI on shares and embeds

This view has reached its Ask AI limit for now

Everyone viewing a share or an embed draws on one hourly limit for the whole share (20 questions unless the publisher set another), which frees up in the next clock hour. The same message shows when the publishing workspace’s AI credits or limits refuse the question. Only the publisher can change either (Ask AI on a share).

This conversation has ended

A shared Ask AI conversation ends 30 minutes after its last answer. Your question is back in the box: send it again to start a new conversation.

Ask AI isn’t available on this view right now

The publishing workspace has turned AI off, or Ask AI on shares and embeds, or its monthly AI budget is used up; or the role of the person who shared the view no longer includes Ask AI on shares, or they are no longer a member of the workspace. Viewers aren’t told which; only the publisher’s administrators can change it (AI controls). When the message is Ask AI is unavailable right now, the workspace’s AI controls couldn’t be checked for a moment; try again later.

Ask AI is unavailable right now

The server couldn’t take the question at the moment; try again later. Ask AI is not available on this server means the server has no AI set up.

For developers: API keys

Every AI feature refuses an API key with a 403: “AI features are available in the DataSquares app; API keys can’t use them.” See Scopes for what a key gets where AI is optional.

Related

• AI in DataSquares — where the AI appears, effort and AI credits.
• What the AI can and cannot do — the limits behind several of these messages.
• AI chat — asking, approving and retrying.
• Seats, plans & AI credits — allowances, packs and auto top-up.