On Monday, Paweł, the owner of a workshop working with a six-person team, hears once again that he should “implement AI”. He wants to know how to use AI in a small business without landing people with yet another place to check. So he maps out 30 days into four stages: from rules and an API key to working with due dates and rules.
Why isn't an account in an AI chat enough on its own?
An account in an AI chat on its own doesn't solve the problem, because it doesn't contain the agreements from your Tasks and conversations. If you have to copy them into a separate window every time, you first have to work out what to copy, then check you haven't missed anything.
Picture a Wednesday at Paweł's workshop. A customer asks about the collection date, a mechanic has added a note to a Task, and part of the agreement was made in a chat. The question “what should I tell him?” asked in an empty AI window doesn't yet have meaningful context. So Paweł asks within the specific matter and deliberately chooses the information the model will see.
There's also a line when it comes to data. Pasting a customer's correspondence into an external tool is a decision to hand its content over to the AI provider, not a minor shortcut in your work. In Task Manager, application data sits on your server, but when you use AI, the query together with its context is sent to the model provider. Paweł tells his team before the first question: “We keep the application's data on our own server, but the question and the supplied context go to the AI provider.”
Week 1: how to start using AI in a small business?
Start with the rules, the person responsible for configuration, and a single API key. In the first week you don't need to give the whole team access to AI: the administrator can enable access to AI in Tasks, Chats and the AI Assistant window separately.
On Monday, Paweł writes down three decisions. Which matters are allowed to be shown to the model? Who is allowed to use it? Can correspondence with customers be a knowledge source? In Task Manager, mail as an AI knowledge source is disabled by default; the administrator can enable it. Mail itself belongs to the Pro plan. Since the workshop doesn't yet have agreed rules on customer emails, Paweł leaves this setting switched off.
Next, he chooses a provider: OpenAI, Google Gemini, Anthropic Claude or xAI Grok. He gets the API key - a string of characters that lets the application use his account with the provider - from his own account. If you're not sure which model to start with, read how to choose an AI provider for your business.
In Task Manager, Paweł opens Settings → AI, chooses the provider and pastes in the key. He uses the refresh button to fetch the list of models available on his account, picks one, and saves the settings. The application tests the connection and shows any warnings. The saved key is encrypted in the database and never appears in full afterwards - even in the Standard plan.
The bill for model usage is separate from the program licence: you pay the provider directly, with no margin added by Task Manager. Early on, also take a look at the Advanced section. The maximum response length defaults to 4096 tokens - a token is a unit of text used to bill the model's work - and longer responses may cost more. Attachment summaries are set to Manual mode by default, so they're only created after clicking “Create summary”. Read more about what gets sent to the AI provider before rolling the feature out to your team.
Week 2: ask within the Task, not in an empty window
In the second week, use AI where the context lives: in a Task or a chat. The AI Assistant can reply within the current Thread, and “Organise” mode helps you read its messages without changing the originals.
On Thursday, Paweł is back after two days away from the workshop. In the Task “Gearbox replacement - Nord”, 23 messages are waiting. He selects “Organise” → “Summarise”: he gets 1-2 sentences per message, starting with the batch of the 8 most recent ones. If he needs older ones, he pulls in another batch with the “Organise older messages (N)” button. The result only exists for this session, and the original messages remain untouched.
When he wants to double-check the arrangements, he clicks “AI Assistant” in the reply field. His question and the answer are private; in the thread he'll see the “FOR YOU ONLY” label. He can also ask for a draft reply or a checklist to tick off. But remember the boundary: a checklist from a private AI Assistant reply in the thread is visible only to its author, and there are no signatures there. A checklist in a message available to the team has a shared state for everyone who can see it, with a signature showing who ticked each item.
Kasia, who is writing a reply to a client in a Task, can click “Enhance text” in the editor. The feature improves the language and keeps the layout of paragraphs or lists; clicking the undo notice restores the previous version. Before sending, Kasia still checks the part numbers and the due date. Below messages processed by “Organise”, the programme reminds you that AI can make mistakes. A summary is a shortcut for checking, not a record of agreed terms.
AI Assistant reads text attachments from the available thread, but it doesn't read the old DOC format or a scanned PDF without a text layer - there's no OCR, i.e. text recognition from an image. If the workshop receives scans, Paweł has to pass on the needed information as text or use a document with a text layer.
Week 3: the reminder should stay tied to the Task's due date
In the third week, link reminders to Task due dates. When you change a Task's due date, the reminder attached to it will move along with it; you don't need to give a new date in a separate command.
On Tuesday, Kasia writes in the Task “Gearbox replacement - Nord”: “remind me the day before the deadline about picking up the car”. The Assistant shows the date and time, but doesn't save anything yet. The reminder is only created once Kasia clicks “Yes, do it”. If the Task has no due date, the AI Assistant will say so and suggest a reminder for a specific date.
Kasia can check the record with the command “show my reminders”; she'll see a list with due dates and IDs. The limit is 50 active reminders. When an AI reply claims to have carried out an action that actually failed, Task Manager displays a “Correction” block underneath it. So after an important command, check the result instead of trusting the wording of the reply alone.
Paweł also enables “Daily task digest” in Settings → Profile. This is one notification a day at a chosen time, 8:00 by default, with Tasks assigned to him for today and overdue ones. If you want to receive an e-mail while the app is closed, the server needs CRON configured - the recurring job scheduler. The notification within the open app works without this; if CRON has never started, the programme warns you when setting the reminder. You'll find more examples in the description of the AI Agent.
Week 4: rules and company knowledge require Pro
Spend the fourth week on Automations and the AI knowledge base, if you're on the Pro plan. On Standard, the AI Assistant still works in Tasks, Chats and its own window, and you point it to knowledge sources within the conversation; there's no Automations or indexed AI knowledge base there.
On Friday, Paweł takes on one recurring issue: filing invoices. He has a mailbox connected in the Mail module on Pro, so he sets up a rule that, on receiving an e-mail, checks the attachment and the AI's assessment, then saves the attachments to the Invoices folder and shows a notification. He writes it in three parts: “WHEN new e-mail, IF it has an attachment and AI assesses it as an invoice, DO save attachments to the Invoices folder and show a notification”. The “AI assesses” condition uses, among other things, the sender, subject, a snippet of the content and the attachment name; it doesn't read the inside of the file. If it isn't sure, it answers “no”. That's a good reason to check the history of skipped events rather than assume every invoice will land in the folder.
Paweł can describe the rule to the AI Assistant in a plain sentence. He'll get an unsaved, disabled draft to refine. The “Test” button shows what would have happened on the latest matching events, but doesn't carry out or send anything. Only after reviewing the draft does Paweł save it. “History” keeps executed, skipped and failed runs for 30 days.
The “AI assesses” condition and the “write message via AI” action query the model on every matching event, so each such request adds to your bill with the provider. Task Manager checks the “AI assesses” condition after the other conditions; the administrator can also set a daily limit on AI requests from Automations, 200 by default. Once it's used up, rules requiring AI are skipped for the rest of that day, with an entry in the history.
On Pro, the AI knowledge base updates in the background once a day by default; the administrator can disable the automatic update and refresh it manually with the “Update AI knowledge base” button. It covers Task and Chat summaries plus the content of Notes and Files; Mail isn't included in this index. The index is a set of snippets prepared for searching out answers. Notes and Files keep their permissions, so you won't see a document in a reply that you're not allowed to open. If you want to separate this stage from working with a single thread, see how the company's AI knowledge base works.
The 30-day plan comes down to four decisions
First you decide who uses AI and what data they can give it. Then you check the AI Assistant's replies within a Task, set a reminder tied to a due date and - if you're on Pro - test a rule and the AI knowledge base.
| Week | What you do | Who's responsible | Option |
|---|---|---|---|
| 1 | You set the rules, add the key, choose the model and permissions | Administrator | Standard / Pro |
| 2 | You check “Organise”, private questions in the thread and text editing | People with AI access | Standard / Pro |
| 3 | You set a reminder tied to a due date and the daily task digest | Everyone, for their own matters | Standard / Pro |
| 4 | You test the rule and decide whether to enable the AI knowledge base | Administrator and people with access to Automations | Pro |
If in the second week summaries miss important details, don't automatically move on to rules. Go back to the original messages and specify which thread or document you're asking about. The AI Assistant in its own window lets you choose sources with the “Sources” button; when there's too much material, it tells you what it left out. That kind of information is more useful than a reply that sounds confident but leaves out half the story.
How much does AI cost in Task Manager?
You pay separately for the Task Manager licence and for usage with the AI provider. The licence is perpetual, with a one-off fee based on the number of seats and 12 months of updates included; renewing updates later is optional.
| Plan | Number of people | Standard | Pro |
|---|---|---|---|
| Starter | up to 8 | 690 zł | 1 990 zł |
| Classic | up to 20 | 1 490 zł | 3 990 zł |
| Max | up to 35 | 2 290 zł | 5 990 zł |
Paweł doesn't buy a separate AI licence for every mechanic: he uses his own key and pays the provider for actual model usage. He still needs to factor in the cost of AI usage and his own hosting, since Task Manager runs on the client's server. When comparing plans, check the current Task Manager pricing instead of treating the licence cost and the model cost as a single fee.
Tomorrow, pick one matter where the team genuinely needs context - say, a Task with arrangements about picking up a car. Agree who can ask AI about its content, choose a provider and set up the key; spread the remaining stages over the following weeks. If you run a team of 3 to 35 people and want to keep your app data on your own server without a per-user subscription, watch the Demo.
Frequently asked questions
Where should you start when implementing AI in a small business?
Start with rules for data and access, then set up a single API key of your own. Run your first attempts on a specific Task or chat, so you can check the response against the original agreements.
Does Task Manager work without AI?
Yes, Task Manager works without AI configured. An account and key with a model provider are only needed once you want to use the AI features.
How much does AI cost in Task Manager?
Costs are split between a one-off Task Manager licence and usage paid directly to the AI provider. Task Manager doesn't add a margin to the provider's bill; you set the maximum response length in the AI tab, and the administrator sets the daily limit of AI requests from Automations in Settings → AI (Automations itself requires the Pro plan).
Is my data sent to the AI provider?
Yes, when you ask AI a question, the query and the supplied context are sent to the chosen model provider using your key. Application data stays on your server, and Mail (the Mail module in the Pro plan) as a source of AI knowledge is disabled by default and is enabled by the administrator.
Which AI features require the Pro plan?
Automations and the AI knowledge base require the Pro plan. Pro also includes Mail and Files, which can be set as knowledge sources; the AI Assistant in Tasks, Chats and its own window is also available in Standard.
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.
