Tuesday, 9:40. Ola Nowak receives a task called “Proposal for client Nord”, due on Friday, with a three-sentence brief. She spends a quarter of an hour going through old messages, although she could start differently: show the AI Assistant the relevant materials and ask it for a plan, checklist or email draft.
The first half-hour is spent piecing together the context
The first half-hour of working on a task can be spent simply searching for the relevant details. You need to check what the client bought before, who spoke to them and which version of the scope is current.
Ola is not yet struggling to write the proposal. First, she is trying to establish whether Nord approved an additional language version, what deadline was mentioned during the conversation and where the latest quote is stored. A general-purpose AI chat does not know this information until you paste it in.
In Task Manager, you can ask the AI Assistant a question within a task thread. You can also open a separate AI Assistant window and select materials related to the matter. The AI Assistant can only see content you have access to. It does not replace an organised project, but it shortens the path from agreed details to an initial draft.
The AI Assistant only knows the context you have access to
The quality of the answers depends on the content in the current thread and the sources you point to. In the Standard variant, the AI Assistant doesn't search the whole company's materials on its own; in Pro it also selects matching fragments from the AI Company Knowledge base, but only from materials you have access to.
Ola opens a task or chat and switches on AI Assistant mode. The question and answer remain invisible to the other participants. The answer stays in the conversation with a “JUST FOR YOU” label, so Ola can come back to it later. Read more about this feature on the page explaining how AI understands the context of a conversation.
In a separate window, Ola selects the materials she wants the AI Assistant to read. She can include her tasks, Calendar, the entire Workspace, a specific task, chat or note, files (a folder or individual file) and her own mailbox - the Mail and Files modules are part of the Pro version. A bar shows how much material fits within the context, and the AI Assistant tells her what it did not take into account.
Every AI Assistant thread has its own history, name and set of sources. It's private even from the administrator. Mail as a knowledge source remains disabled by default, because once enabled, its content is sent to the external AI provider. The AI Company Knowledge base is part of the Pro variant.
The “create a plan” prompt turns your materials into an initial draft
The “create a plan” prompt helps you get started on a task without manually copying all the agreed details. Treat the result as a draft sequence of actions to review before sharing it with the team.
Ola opens the AI Assistant window and selects the “Proposal for client Nord” task, the Sales chat and the note containing the scope of services. At 10:05, she enters:
Create a plan for preparing the proposal for Nord. Base it solely on the selected task, the Sales chat and the note containing the scope. Flag any information I need to check.
In this scenario, the response might look like this:
- Confirm the website scope and language versions.
- Check the delivery date recorded in the conversation.
- Prepare the quote based on the note containing the scope.
- Verify the person responsible for approval at Nord.
- Write a message containing the proposal and response deadline.
The plan is not a decision or a finished proposal. It is an organised draft based on the materials provided. If some sources do not fit within the context, the AI Assistant will identify the omitted materials rather than leaving the limitation unmentioned.
The “create a checklist” prompt turns the plan into action
The AI Assistant can turn a task description into a checklist you can tick off. This prevents the plan from becoming yet another long response that no one reads.
On Wednesday at 8:50, Ola opens the “Website project” task and enters:
Turn the task description into a checklist I can tick off. Arrange the items in the order they should be completed.
The AI Assistant returns a list, and you tick off completed items directly in the thread. The progress bar below the list advances. Ola reviews and edits the items, then moves them to the shared task description.
When the list is intended to guide several people's work, you move it to the task description. The To-do list in the description is shared with the team, works without opening the editor and shows the name of the person who ticked off each item. On Thursday at 15:20, Marta ticks off “quote checked”, and the rest of the team can see the change without asking in the chat.
See also checklists created by AI and an example of how a checklist organises a team task.
The “draft an email” prompt still requires your review
The AI Assistant can draft a message based on the specified agreed details, but it will not send it automatically. The email is only sent from your personal inbox after you confirm it.
On Wednesday, the quote for Nord is ready. Ola asks:
Draft an email to the client Nord. Confirm the scope of the quote and the project delivery date, and ask for approval by Friday. Do not add any information that is not in the sources.
The AI Assistant might suggest:
Hello, I am sending the agreed scope of the quote for the new website. The project includes the elements described in the attached summary. Please confirm the scope by Friday so that we can keep to the stated schedule.
Ola checks the address, date and scope. If “2 days for implementation” appears in the draft, she returns to the source to check whether that figure was actually mentioned. She can request a correction in a single sentence, but responsibility for the content still remains with her.
The Mail module is included in the Pro plan. You can learn more about sending messages through the AI Assistant on the sending emails with AI page and in the description of the Mail module.
Read a thread with 40 messages in Expert mode
Expert mode helps identify risks, contradictions and missing details in a long conversation. Organise then displays a temporary view of the content, while the original messages remain unchanged.
| Mode | What it does | When to use it |
|---|---|---|
| Original content | Displays messages without changes | When you need the exact wording of the agreed details |
| Polish | Shortens the same content and rewrites it in the first person | When messages are chaotic or full of typos |
| Summarise | Condenses each message into 1-2 sentences | When you want to quickly reconstruct the conversation |
| Expert | Rewrites the content professionally and adds an “Expert note” line | When you are looking for a risk, contradiction or missing detail |
For the Nord quote, Expert mode might flag that one message gives payment terms of 30 days, while another gives 14 days. This is a prompt to investigate, not an automatic amendment to the contract.
Organise processes conversations in batches of 8 messages. The program does not save the organised versions after the session ends. If you export a message while reading the AI version, the file will contain the version currently displayed, with a note indicating the mode used.
The AI Agent takes on tasks involving several steps
The AI Agent is an advanced mode of the AI Assistant: after receiving a complex instruction, it proposes an action plan for you to approve before it is carried out. This mode is designed for complex instructions that cannot be completed in a single response.
On Friday, Ola needs a summary of the agreed details for the Nord quote. Instead of reviewing the threads one by one, she enters a complex instruction: collect the agreed details from the task and the Sales chat, then prepare a short summary. The AI Agent proposes an action plan with the next steps.
The plan only starts after you approve it and operates within your permissions - the AI Agent does nothing that you have not approved yourself. You can find more examples in the description of the AI Agent for multi-step tasks.
Reminders set in the AI Assistant always work when the application is running. Sending emails in the background while the application is closed requires a recurring CRON job on the server; the program warns you about this when you set a reminder. CRON is a recurring job run by the server at a scheduled time.
AI ends where responsibility for the outcome begins
The AI Assistant prepares drafts and organises content, but it does not take responsibility for decisions. Check numbers, dates, addresses and names before using the response.
The AI Assistant advises, but does not decide. A response may sound confident even when the source material is incomplete. This is why you should compare the draft with the agreed details before sending a quote.
The AI Assistant can read specific formats. It supports TXT, CSV, MD, JSON, XML, HTML, DOCX, XLSX and PDF. It can also analyse images if the selected model supports this capability. The program does not have OCR, so a scanned contract without a text layer will not be read like a standard document. It also cannot read old binary DOC files.
The AI Assistant does not know about details you have not recorded. If the client changed the deadline during a phone call, add a sentence to the thread or a note. The model cannot reconstruct a conversation that is not in the sources.
The program also works without AI. Tasks, to-do lists, Calendar and Notes do not require an account with a model provider. The administrator enables AI separately for tasks, chats and the AI Assistant window.
The question and the necessary context are sent to the selected provider: OpenAI, Google Gemini, Anthropic Claude or xAI Grok. Your company uses its own API key and pays the provider directly, with no Task Manager markup. Application data remains on the company server, but this does not mean that queries sent to the active AI stay on that server.
Tomorrow, start with one overdue task
Tomorrow at 9:00, choose a task that has been sitting there for two days and ask the AI Assistant for a plan to get started. You do not need to change how the entire team works straight away.
First, the administrator selects a provider under Settings → AI and enters the company’s own API key. They then download the list of models and select one. Once the configuration is saved, the AI Assistant automatically appears for other users, usually within about a minute. The process is described in the guide to using your own AI key in your company.
Select a chat or note linked to the task and type: “Put together a plan for getting started on this task. Use only the sources I’ve selected.” You’ll still need to do the task, but you’ll start with a structured outline instead of spending fifteen minutes searching through your inbox. You can try this out in the Demo.
Frequently asked questions
Can other task participants see my question to the AI Assistant?
No, a question asked in AI Assistant mode within a thread and the response labelled “ONLY FOR YOU” are private. A task saved from the prepared form and an email sent after confirmation become standard actions in the application.
Can the AI Assistant see other people's tasks?
The AI Assistant can only see content you have access to. Notes, your own mail and conversations with the AI Assistant also remain private from the administrator.
Can the AI Assistant send an email or create a task without my knowledge?
No, an email is only sent after you confirm it, and a task is only created when you save the form. The AI Assistant can complete the form fields, but it cannot bypass this step.
Where does my data go when I ask the AI Assistant a question?
Application data is stored on your server, while the question and the necessary context are sent to your chosen AI provider using your API key. Mail as a knowledge source is disabled by default and must be enabled by an administrator.
What should I do if the AI Assistant does not respond?
Check the selected provider, API key and model in AI Settings. If the configuration is correct, the cause may be an API key limit or no internet connection.
Sources
The Task Manager Team - we write about teamwork, AI in business and keeping tasks organised, using examples from Task Manager, software installed on your own server. Task Manager has been developed by Blue Creation Group (Piotr Łącz) since 2013 and is used by organisations including Neonet, the Military University of Technology and Polpharma - read customer reviews. We check our articles against the user guide and the latest version of the software.
