Sistava

How to Create Your Own Personal AI Assistant, Step by Step

How-to — by Mahmoud Zalt

Create your own personal AI assistant: connect inbox and calendar, write a one page brief, set the approval line, and let it run daily.

Most people asking this question have already tried the easy version. You opened a chat window, asked it to write an email, got something decent back, and then closed the tab and went right back to doing everything by hand. Nothing changed, because a chat window does not know your calendar, does not see your inbox, and does not do anything while you sleep.

Creating your own assistant means closing that gap. Not a smarter chat window. A worker with your accounts connected, written instructions it follows every day, and a clear rule about which actions it takes on its own. Once those three things exist, the difference is immediate and a little strange. Work starts getting done in the hours you are not looking.

On Sistava, you do this by hiring an AI Employee whose role is Personal Assistant. She is called Alice. Hiring her creates a private worker inside your workspace with her own memory, her own connected accounts, and her own schedule. She is not a shared chatbot everyone else is also talking to. What you tell her stays with her, and what she learns about you gets used the next time you ask for something.

At a Glance

4
Things to set up: accounts, brief, boundaries, schedule
30 min
Typical time from signup to a working assistant
0
Lines of code you have to write
$25
Starting price per month on the Starter plan

What does creating your own AI assistant actually involve?

It involves four things, in this order: connect the accounts it needs, write it a brief, set the boundary between acting and asking, and give it a daily schedule. Everything else people talk about is either a detail inside one of those four or a distraction from all of them.

The order matters more than it looks. Connecting accounts first means the brief you write can refer to real people and real meetings instead of hypotheticals. Setting boundaries before the schedule means nothing runs unsupervised until you have said what unsupervised means. People who skip straight to the schedule usually end up turning the whole thing off in week one.

How do you write the brief your assistant runs on?

Write it the way you would write a handover note for a competent human starting Monday. One page, plain sentences, no formatting tricks. Cover five things: what you do, what you are trying to get done this quarter, who the important people are, how you want to sound in writing, and what you never want to be interrupted for.

Be specific in a way that feels almost rude. Not "be professional" but "short sentences, no exclamation marks, sign off with just my first name." Not "handle my email" but "anything from my three biggest clients comes to me the same hour, everything else waits for the morning brief." Vague instructions produce vague work, and you will feel that on day two.

The brief is a living document, not a form you fill in once. Every time your assistant gets something wrong, the fix is usually one sentence added to the brief rather than a lecture in chat. After three or four weeks most people have a page that reads like a genuine description of how they work, which is oddly useful even for the humans they work with.

Where do you draw the line between acting and asking?

Draw it at consequence. If an action is reversible and invisible to other people, let her do it alone. If it is visible to someone else, costs money, or would be embarrassing to undo, it needs your yes. That single rule covers almost every case you will hit in the first month.

In practice this means she sorts, labels, drafts, researches, summarises, and books internal time without asking. It means she does not send an email to a client, accept an invitation on your behalf, or spend anything until you have looked at it. Approvals arrive as a short card with the exact action and the exact text, and you tap yes or no. Most people spend under five minutes a day on this.

ActionAlone or asks firstWhy
Sort and label incoming mailAloneReversible, nobody outside sees it
Write the morning briefAloneIt is for you and only you
Research a company before a callAloneReading, not acting
Draft a reply to a clientAlone (draft only)A draft is not a sent message
Send that replyAsks firstSomeone else reads it and it cannot be taken back
Book a meeting with an outside personAsks firstIt occupies another person's time
Block your own focus timeAloneIt is your calendar and you can move it
Anything involving money or contractsAsks first, alwaysConsequence you cannot reverse with an apology

A worked example: Priya's first two weeks

Priya is a freelance user experience consultant in Lisbon with six active clients. Before she set anything up, she was getting roughly ninety emails a day and losing the first ninety minutes of every morning sorting them. Two of her clients were slipping to a next day reply, which she knew was quietly costing her renewals.

She connected Gmail and Google Calendar on a Sunday evening. Her brief was eleven sentences long and named all six clients, the two she could not afford to keep waiting, her rule that she does no calls before eleven, and one line about tone: no filler openings, get to the point in the first sentence. She set the approval line so that nothing leaves her outbox without her tapping yes.

Week one was corrections. Her assistant flagged a newsletter as urgent because it came from a client's domain, and Priya added one sentence to the brief about that. It wrote a reply that opened with a compliment, and Priya deleted it and said why. By Friday the drafts needed edits about a third of the time instead of every time.

Week two was quieter. The morning brief was waiting at seven, listing the four things that actually needed her and summarising the rest in a paragraph. The two clients who used to wait were getting same day replies, because the drafts were sitting ready and approving one takes fifteen seconds. Her ninety minute morning became about twenty. She did not save ten hours a week. She saved around five, and she stopped starting every day annoyed, which she says mattered more.

At a Glance

90 min
Priya's old morning inbox routine
20 min
Her routine by the end of week two
11
Sentences in the brief that made it work
0
Emails sent without her approval

How do you set one up today?

Follow these five steps in order. Give yourself half an hour and do it in one sitting, because the pieces only make sense together.

  1. Hire the assistant — Create a workspace and hire the Personal Assistant role. That gives you a private AI Employee with her own memory and her own connected accounts, rather than a shared chat session that forgets you between visits.
  2. Connect your inbox and calendar — Sign in to Gmail or Outlook and to your calendar through the normal permission screen. This is the step people delay and it is the step that makes everything else possible. Nothing useful happens while she is guessing at your week.
  3. Write the one page brief — Who you are, what you are working on, who the important people are, how you want to sound, and what you never want interrupted for. Plain sentences. Twelve of them is plenty. Paste it in and let her read it.
  4. Set the approval line — Tell her what she may do without asking and what always needs your yes. Start tight: drafts yes, sending no. You can loosen it in week three when you have seen the quality for yourself.
  5. Give her a daily schedule — Pick a time, usually thirty minutes before you start, when she reviews the inbox and calendar and writes your brief unprompted. This is the switch that turns a tool into an assistant. Everything before this step is setup, and this step is the one that pays.

What should you never hand to a personal AI assistant?

Anything where being wrong costs more than being slow. Your assistant is fast, tireless, and confident, and confidence is exactly the problem in the places where a human would hesitate. Keep these on your own desk, permanently, not just at the start.

One more thing worth saying plainly. The assistant is only as good as the honesty in your brief. If you write down how you wish you worked instead of how you actually work, she will optimise for the fictional version and you will spend a month wondering why it feels off. Write the real one.

FAQ

Do I need to know how to code to create a personal AI assistant?

No. Creating one is signing in to your email and calendar, typing a page of plain instructions, and choosing which actions need your approval. There is no model to train, no API key to manage, and nothing to host. If you can write a handover note for a new colleague, you can do this.

How long does it take before it is genuinely useful?

The setup is around thirty minutes. Usefulness arrives in stages. Day one you get a working morning brief and drafts that need editing. By the end of week one, after you have corrected the obvious misses, roughly two thirds of drafts go out with small edits or none. Week two is when most people stop thinking about it as a project.

What does my assistant do while I am asleep?

Whatever you scheduled and permitted. Typically she reads overnight mail, sorts it, drafts replies to the routine ones, checks your calendar for conflicts and missing prep, and has a brief ready before you wake up. She does not send anything to another person unless you gave her that permission explicitly.

Can I create more than one assistant for different parts of my life?

Yes, and some people do, usually one for work and one for personal admin. It is not required. A single assistant with a good brief handles both fine, and one memory that knows your whole week is often better than two that each know half of it. Start with one.

What happens to my email data?

It stays in your own workspace, is encrypted in transit and at rest, and is not used to train any model. You choose which accounts she can reach, you can exclude specific senders or labels from her view, and you can disconnect an account at any time and she loses that access immediately.

How is this different from just using a chat assistant every day?

A chat assistant waits for you to open it and forgets you when you close it. A personal AI assistant has your accounts connected, a written brief she follows, permission rules that let her act, and a schedule that starts her working before you do. The difference is not intelligence, it is standing instructions and access.

The whole point of creating your own assistant is not to collect a new tool. It is to stop spending the first ninety minutes of the day on work that never needed you. Connect two accounts, write one honest page, decide what she may do alone, and let tomorrow morning be the test.