Performing Long Tasks
This feature is intended for the desktop application. For regular questions and short tasks, you can continue using the web version.
With the desktop application, you can have an extensive task executed in the chat. The AI divides the task into steps, completes them one by one, and can read, create, and modify files in a folder you choose.
This is useful if you want to:
- compare multiple documents;
- merge information from different files;
- create an analysis and a final document;
- perform a task consisting of many consecutive steps.
Starting a Task
- Open the desktop application.
- On the start screen, choose the folder containing the files you want to use, or approve it with Add folder….
- Open a chat and describe the desired end result as clearly as possible.
- Specify the source folder, the desired output folder, and the files the AI should produce.
- Send your task and keep the desktop application open while the task is running.
Make a plan first
Click the Plan button below the input field (the icon with a list and check marks; hover over it for an explanation) or start your message with /plan to make a plan together first. The icon is colored while plan mode is on. The assistant then only reads along, asks questions if needed and changes nothing.
The proposal appears as a card with a goal, steps, success criteria and risks. Then choose:
- Execute plan to start;
- Edit to change, add, remove or reorder steps or criteria;
- Discuss further to refine the plan first.
Working with a goal
An executed plan becomes the goal of the chat. You can also set a goal with /goal followed by your goal, or with the Set as goal flag below an answer.
Above the input bar you see the goal, its status (active, paused or reached) and which success criteria have already been met. This is where you manage the goal:
- Pause stops the work. The message then shows "Goal paused" and queued messages stay where they are;
- Resume in the goal row above the input field lets the assistant continue;
- Edit goal changes the goal and criteria;
- Clear goal removes the goal.
While a goal is active, the assistant keeps working on its own for a few more rounds after finishing a step. The goal is saved on this computer, even after you restart the app.
Steering and follow-up messages
While a task is running, you can already type your next question:
- press Enter to steer the running task; the AI applies your addition at a subsequent step;
- press Tab to add the message to the queue. You can also do this with the arrow next to the send button.
Queued messages appear above the input field. You can edit, reorder or remove them. As soon as the task finishes successfully, the next message is sent automatically, even if you have moved to another page in the meantime. If you have another chat open, this happens as soon as you return to this chat. If the task stops or fails, or the goal is paused, the queue stays put and you choose Send now yourself.
In Settings > Tasks, use Default for Enter during a task to choose what Enter does: Steer or Queue.
Reasoning
Below the input field you find the Reasoning button (the icon with a head and a gear). Hover over it to see the current choice, for example "Reasoning: Default". Click it to choose how thoroughly the model thinks: Default, Low, Medium or High. Higher is more careful, but slower and more expensive. If a choice other than Default is set, the icon is colored. The choice is remembered per chat and only appears for models that support it.
Monitoring Progress
Choose the Tasks and activity button on the right of the top bar. The File changes, Commands, Tasks and activity, Files and Browser buttons sit there side by side as icons, with the name of the active workspace next to them (for example "Documents"). Hover over a button to see its name, or over the folder name to see the full path. A panel only opens beside the chat when you choose its button. The panel shows the available tasks and activity for the latest response and updates during execution. Progress remains available afterwards; unfinished steps are marked as incomplete. Click the same button again, close the panel or press Escape to close it.
Sometimes more steps are needed than initially expected. You will receive a notification where you can allow the AI to continue or stop the execution.
Running tasks
The task icon in the toolbar shows how many tasks the assistant is currently running. Click it to see the status per chat (working, waiting for you, done or error) and open a chat with one click.
The start screen
When you open the app or choose New Chat at the top of the sidebar, you see a calm start screen instead of the dashboard with tiles. Under the heading "What do you want to work on?" you find a folder button and the input bar; your default workspace, for example "Documents", is already selected.
- With the folder button you choose one of your projects, your default workspace (marked "default") or another approved folder. Add folder… approves a new folder.
- The chat you start works in the chosen folder or project. Your choice stays selected for your next new chat.
The web version and the mobile app keep the dashboard with tiles.
The desktop application is made for chatting with the assistant. That is why there are no separate modes for images, text to speech, documents or voice chat. If you want an image or a document, simply ask for it in the chat; the assistant creates it with its tools. Voice assistants cannot be started in the desktop application: on the Assistants page their Chat button is disabled. Use them in the web version. Dictating with the microphone in the input field does work.
Chats in the sidebar
On the left of the desktop application, a sidebar lists your folders with their recent chats. Your desktop projects come first, then your workspaces. Chats without their own folder appear under the default workspace, for example "Documents". Pinned chats are at the top.
- Click a folder name to expand or collapse it. Show more shows more chats of that folder.
- Hover over a folder and click + to open the start screen with that folder selected. New Chat at the top starts a new chat.
- Each chat shows its status: a spinning icon while the chat is working, an orange dot when it is waiting for you (for example a command that needs your approval), a blue dot when it is done but you have not opened it yet, and a red icon for an error. Opening the chat clears the blue dot.
- At the bottom you find the navigation (Search and actions, Transcription, Assistants, File management, Groups and, if you have access, Admin), help (Kennispunt) and your profile. Your history is the chat list itself.
- Collapse sidebar next to New Chat collapses the sidebar; the navigation then moves back to the top bar. Expand sidebar on the left of the top bar opens it again.
Notifications and sleep
Is the desktop application working on a task while you do something else? Then your computer shows a notification as soon as the task is done, fails or needs you. The latter happens, for example, when a command is waiting for your approval, or when the task has used many steps and asks whether it may continue. A reached goal also triggers a notification.
Click the notification to go straight to the right chat. If the chat is already open in front of you, you will not get a notification. On macOS, the app icon in the Dock shows how many tasks are waiting for you.
While a task is running, the app prevents your computer from going to sleep.
You turn both features on or off under Settings > Notifications and sleep, with Notify me when a task finishes or needs me and Prevent sleep while tasks run. Both are on by default.
Stopping and resuming a task
Use the stop button next to the selected AI model to stop a running task. The AI will then not perform any new steps. Changes already made to files are not automatically undone.
You can also stop a running task with Cmd+. (Mac) or Ctrl+. (Windows), see Command palette and keyboard shortcuts. To undo changes to files, use File changes, see Local File Access.
If you stop a task and immediately send a new message, you briefly see "The previous task is still stopping". Your message starts automatically as soon as the previous task has actually stopped.
With Resume task, a stopped task continues where it left off. Read actions and actions that had not started yet, such as a command that was still waiting for your approval, are simply run again. For a command, the app asks for your approval again.
Was an action running when you stopped, and is it unclear whether it was carried out, such as sending an email through a connection? Then the app shows that action with Resume anyway and Cancel. If you choose Resume anyway, the action is not repeated: the assistant is told that the outcome is unknown and first checks the current state.
Approving a folder allows the assistant to read and change files in it. Only commands and starting local connections ask for approval each time. For commands, you can change this per folder, see Running local commands safely.
Run extensive or new tasks first on a copy of your files. Check the results before replacing the original files.
Example Task Without Internet
Use only the files in the
sourcefolder. Read all documents and create a summary per document in theresultfolder, a comparison of the main similarities and differences, and a brief final recommendation. Also create an overview of the used filenames. Do not change anything in thesourcefolder.
Tips for a Good Result
- Specify exactly which files the AI may use.
- State what may and may not be changed.
- Indicate in which folder the results should be saved.
- Describe which final files you expect and for which audience they are intended.
- Check intermediate notifications if the task takes a long time.
- Keep in mind that extensive tasks may require more time and AI usage.
In a narrow desktop window, the five panel buttons appear above the chat. The active workspace remains visible. Panel content opens over the chat, keeping the input full width when the panel is closed. Close with Escape, the close button or a click outside the panel. Escape and the close button return keyboard focus to the corresponding panel button.