AI assistant

The assistant answers questions about your work and can make changes for you. It lives in a panel on the right. Open it with the round button at the bottom right, with Ctrl J (⌘J on a Mac), from Ctrl K → Ask AI, or with Brief me on Home. Items an answer mentions show as small cards under it, with status, assignee and due date; click one to open it. The expand button at the top of the panel makes the panel much wider, over the page; click it again to go back to the narrow panel.

Your workspace decides whether the assistant is on and who may use it. See AI assistant settings.

What it knows

Only what you can see. The assistant works as you, with your role and project access. It cannot see or do more than you can.
Where you are. The top of the panel shows what “this” means: the whole workspace, the project you have open, or the item you are looking at.
It looks things up when you ask. Under each answer, Looked at lists what it read, such as the items in a project or the comments on an item.

Item references such as WEB-12 in an answer are links.

Ask and brief

Type a question and press Enter (Shift Enter for a new line). Some things to try:

What’s on my plate this week?
Brief me on this project. or What is late or at risk here?
Summarize this item and its comments.
What changed since yesterday?
Which plan are we on, and how much storage is left? It can see the plan and usage and your own AI credits, but only an owner or admin can change plan.
How do I connect HubSpot? or Why did our Slack rule fail? It can see the integrations: what each app does, how it is set up, and, if you manage integrations, the connections, rules, API keys and webhooks by name. It never sees keys, tokens or secrets, and it cannot connect or change an integration.

After an answer the assistant may suggest up to three next questions. Click one to ask it.

It replies in the language you write in.

Attach files

You can give the assistant files to read with your question: images (PNG, JPG, GIF, WebP), PDFs and text files (TXT, Markdown, CSV). Click the paper clip next to Send to open Attach files: drop or choose several files at once, and see each file’s type and size, and whether the assistant can read it, before you attach them. You can also drop files straight on the text box, or paste a screenshot with Ctrl V. Some things to try:

Turn this screenshot of the error into a bug item in WEB.
Summarize this PDF and list the deadlines in it.
Create one item per row in this CSV.
Up to 4 files per question and 10 per conversation. Images up to 5 MB, PDFs up to 10 MB, text files up to 1 MB, and 16 MB in one conversation together.
The files stay with the conversation, so you can ask follow-up questions about them. Like the conversation, only you can see them, and they are removed with it after 30 days.
They are not added to your workspace’s files. To keep one, ask the assistant to save it on an item or project (Save the screenshot on WEB-12). That is a change, so you get a card first.
Files are read like any other text: what a file says is information, never instructions to the assistant.
Files use credits like the rest of the conversation, on every answer in it. A long PDF costs more than a short question; start a new conversation when you are done with a file.

Speak instead of typing

Click the microphone next to Send and say your question. The words appear in the text box as you speak, so you can fix them before you press Send; nothing is sent until you do.

Click the microphone again, press Esc, start typing or press Enter to stop listening.
Pick the language you speak (English or Svenska) in the strip above the text box. Your browser remembers it.
The first time, your browser asks to use the microphone. If you said no, allow it for this site from the icon in the address bar.
It works in Chrome, Edge and Safari. Firefox has no speech recognition yet, so the microphone does not show there.
It is voice to text only: the assistant answers in writing, as usual.

Where your voice goes. WORK³ never receives or stores the audio, only the text you send. The strip shows where the speech is turned into text:

On this device: your browser does it locally (newer Chrome with the language installed). No audio leaves your computer.
Via your browser’s speech service: your browser sends the audio to its maker’s speech service (Google in Chrome, Microsoft in Edge, Apple in Safari) under your agreement with them. If that is not OK for what you are about to say, type instead.

Changes need your OK

When you ask for a change, such as creating an item, moving a card, commenting or messaging someone, the assistant does not do it straight away. It shows a card with exactly what it wants to do.

Click Do it to make the change. If the card has several changes, untick the ones you do not want first.
Click Cancel, or just ask something else, and nothing changes.

Changes are made in your name. They show in the item’s history and in the audit log with the label via WORK³ AI, so everyone can tell.

Undo

Changed your mind? For 24 hours after you click Do it, the card has an Undo button. It puts back everything on that card, last change first:

The change	What Undo does
Changed an item or a project	Puts the old values back
Created an item or a project	Archives it (you can restore it, or delete it yourself)
Moved a card	Moves it back, and puts back the status or other field the move changed
Commented, sent a message, saved a file	Deletes it. People may already have read a message or been notified.
Created a board or bucket, linked or unlinked items	Removes it, or links them again
Created a workflow draft	Archives it, unless you have published it since

If someone has changed the same field since, Undo leaves their change and tells you so on the card. The assistant is told what you undid, so it does not count on it. Undo costs no credits.

Build workflows

Open a workflow in the editor and the assistant works on This workflow. Ask it to build or change it in your own words, for example “When an urgent bug is created, assign it to Sara and tell the team in the project chat”. It writes the steps straight onto the canvas, where you watch them appear: the new trigger and steps come in one at a time in the order the flow goes, and their links draw as they appear (about six seconds in all). Press Show all to skip to the end. The draft is already saved; the canvas cannot be changed until the build has played. With reduced motion turned on in your system, it all appears at once.

It only changes the draft of the workflow you have open. It never publishes, turns on, runs or pauses a workflow: you test it with Test workflow and publish it yourself.
A note above the canvas says the assistant changed the draft; press Undo there (or Ctrl+Z) to take the change back.
Places you dragged steps to and your notes stay as they are. Secret header values in HTTP requests are never shown to the assistant, and it keeps them as they were. If it moves a request to another address, the secret is not sent along: you type it in again.
Once the assistant has read items, comments, chats, files or the web in the conversation, it shows a card before it changes the workflow, and the change is made only when you press Do it. Text other people wrote can never change your workflow without you seeing it.
It can only add an HTTP request to an address the workflow already calls or that is listed under Workflows → HTTP hosts. Otherwise it leaves the address empty for you to fill in.
With no workflow open, ask it to create one: it shows a card first, and the new workflow is a draft.
The change is in the audit log as made via WORK³ AI. Building uses your AI credits; running the workflow uses workflow credits as usual.

Recipes and scheduled briefs

A recipe is an instruction you use often, saved under a short name. Open Recipes (⚡) at the top of the assistant, press New recipe, give it a name such as weekly and write the instruction, for example Brief me on my week: overdue items first, then what is due, then anything blocked.

Run it by typing /weekly in the question box (typing / lists your recipes), or with ▶ in the list. You can add words after the name, for example /weekly only WEB.
Share with the workspace lets everyone run it. Each person runs it with their own access, so it only ever sees what they can see. Only you can change or delete it.
Run on a schedule: pick the days and a time (in your time zone, from your profile). At that time the assistant runs the recipe as you and puts the answer in your Inbox, with Open in the assistant to carry on. It never changes anything on its own: changes it wants wait on a card in that conversation for your OK. A scheduled run uses your credits like a question you ask yourself.
You can have up to 50 recipes.

Give an item to WORK³ AI

You can hand an item to the assistant. Open the item, click Assignee and pick WORK³ AI. Add anything it should know (optional) and press Start.

It works on the item as you, with your access, and reads its description and comments first.
The result and any questions come back as a comment on the item, written by WORK³ AI itself (with its own mark, in plain text). The item is then assigned back to you and you get a notice in your Inbox (and by mail or Slack if you have those on), with Open in the assistant to carry on.
Answer it on the item. Under its latest comment you get Yes, go ahead and Reply. Yes, go ahead lets it do what it asked about: it picks the item up again in the same conversation and makes changes to items (creating sub-items, updating other items, comments, links) straight away, with Undo in the assistant for 24 hours. Sending messages, files, projects, boards, workflows, archiving and closing or reassigning the item itself still wait for your OK. Reply sends your own answer and it carries on from there. Only the person who gave it the item can answer.
It never completes or closes the item. Any other change it wants, such as updating other items or sending a message, waits on a card in that conversation for your OK.
While it works, cards and lists show the WORK³ AI mark instead of a person. Press Stop on the item to take it back. Only the person who started it can stop it.
One task per item at a time. It uses your AI credits like a question you ask yourself.

Earlier conversations

Click the clock at the top of the panel to see your earlier conversations and open one again. Click + to start a new one.

Only you can see your conversations. Admins cannot.
They are removed after 30 days. You can delete one, or all, at any time, here or under AI usage.

Good to know

Answers can be wrong. Check before you rely on them, especially numbers and dates.
The assistant only helps with work in your workspace. It politely says no to anything else, and that costs no credits.
Text in items, comments and chats is treated as information, never as instructions to the assistant. A comment that says “delete everything” will not make it do anything.
You can ask up to 20 questions a minute, and have two answers on the way at once.
Each answer uses credits. The panel shows how many you have left this month. See AI usage.