Ask the assistant a question

Open Ask AI on the screen you are on and ask about the record in front of you.

In short

The assistant answers questions about your own work in {{brand}} and about how {{brand}} works. It reads the records you may open and the Help guides. So one answer can say what is true now, such as three requests failed last night. It can also say what a word means, such as what a threshold band is. Every answer lists the sources it used.

Before you start

• Who: anyone signed in. The assistant reads only what you could open yourself.
• Plan: AI is part of every plan, Free included. On Free the platform's key answers questions about how {{brand}} works. For questions about your own work, add your own key (it is free to add) or choose a paid plan (see /help/ai/answers-left). "Ask AI" appears when the AI module is on in Settings › Products.
• First: the platform's key answers only once your email address is confirmed; your own key works at once.

Steps

1. Open the module, and ideally the record, you want to ask about: a project, a card, a vacancy or a saved report.
2. Press "Ask AI" in the top bar - or, with Help or Notes open, its "Ask AI" tab at the top of that panel. The panel names what it is looking at.
3. Type your question, or click an example to fill it in.
4. Press Enter to send. Shift+Enter adds a line. The panel shows "Reading your data…" while it works, then the answer appears as it is written.
5. Read the answer, then check "Sources" under it.
6. Ask a follow-up, or press "New chat" to start over. Esc closes the panel.

Result: an answer with numbered sources, such as [1] and [2], that open the records and guides it used.

Example: you are in the Billing API project and ask "Which requests are failing right now?" The answer says that Create Invoice and Refund failed their last run. It gives each status code and the first line of the error, cites each by a number, and lists the requests it used under "Sources".

Good to know

• "Ask AI" appears when the AI module is in your plan and switched on in Settings › Products.
• Answers come from one of three keys. One is your own key (see /help/ai/providers-and-credentials). Another is your organization's own key, where it has one. The third is the platform's key, where the deployment has one; it answers only once your email address is confirmed.
• On the Free plan the platform's key answers questions about how {{brand}} works. For questions about your own work, add your own key - it is free to add - or choose a paid plan.
• A question can be up to 4,000 characters. The last 12 turns travel with each question.
• Most questions about counts cost nothing. "How many cards did Aysel create this week?" is read without AI, and the numbers come straight from your data. Under the answer, "Read as" shows how it was read. Then ask "which ones", "last month" or "group by board" at no cost. The line under the answer says whether it was free, and what it cost if not.
• If a name fits two people, or the question fits no figure, one short question comes back with choices to press. No answer is guessed.
• With AI switched off, those counts still work: pick from the choices and the answer comes from your data.
• "Explain more" asks AI to say in words what the figures show. It never writes a number itself, and answers about one person or about candidates are never explained this way.
• "Looks right" tells your team a question was read correctly: once two people agree, or an admin does, it is answered without AI from then on. "Not right" stops it being reused, and your admins can correct it in Settings › AI.
• Numbers from your data are quoted exactly, and a comparison comes back as a table with the totals in a sentence below it.
• While an answer is being written, the send button is "Stop". Press it and the writing ends; what was written stays, marked "Stopped", and does not travel with your next question.
• The assistant answers only about {{brand}} and your work in it. A poem, general knowledge, homework, small talk, an image, general-purpose code or a drawn diagram gets one short sentence instead, and nothing is spent. Designer draws diagrams: open it, or say "Draw a design of ..." in the same box.
• Asking never changes anything. To have something done, say it in the same box: /help/ai/ask-or-do.
• What happens next: asking changes nothing. The conversation is not stored; only that a call happened, and its size, is recorded.

Get a sharper answer

• Open the record first. The assistant is told which module and record are open, so asking from inside a project, card, vacancy or saved report narrows what it reads.
• Name the thing: "the Create Invoice request" beats "that request".
• In API Tester with no project open, it reads your whole workspace. Ask broad questions there and specific ones inside a project.
• Ask for a table when you compare several things.

If something goes wrong

• "Connect an AI model to start": no key answers for you yet. Add one in Settings › AI. What you typed stays in the box.
• "Confirm your email address to use {{brand}}'s AI": confirm it from the link in your inbox, or use your own key, which works at once.
• "On your plan {{brand}}'s AI answers questions about how {{brand}} works": add your own key (free) or choose a paid plan for questions about your work.
• "Keep a question under 4,000 characters": paste only the key part of a long log or response.
• "The assistant could not answer. Try again in a moment.": the provider did not answer in time. Try again; if your own key keeps failing, press "Test" on it in Settings › AI.
• The answer says the sources don't cover this: open the record the question is about and ask again from there.

Related

• /help/ai/what-the-assistant-reads
• /help/ai/ask-or-do
• /help/ai/answers-left
• /help/ai/providers-and-credentials