A personal AI assistant: how to build yours and make it remember

The short version: a personal AI assistant differs from an ordinary chat in three ways — it remembers you between conversations, knows your files and writes first when it is time for a reminder. You can build one in ten minutes: an agent with memory, search and scheduled runs, plus short instructions. Below: how the memory works, and a ready prompt.

AI assistant vs AI agent: the difference

In practice a good personal assistant is an AI agent set up for one person. That is why you build it with the same tools described in the guide to AI agents.

How the assistant remembers

"Remembers your preferences" appears in every advert. Here is how it actually works: the assistant writes important facts as short, dated lines into its own memory document. Before every answer, up to 4,000 characters of that memory go into its instructions: first what is marked as important (your name, how you like to be addressed), then the most recent facts.

# empty

How to build a personal AI assistant, step by step

1. Create a new agent and give it a name — an assistant with a name is easier to talk to.
2. Paste the instructions below into the "System Prompt" field and adjust them to suit you.
3. Switch on the capabilities: "Memory", "Scheduled runs", "Web Search" and "Documents". Anything else as needed.
4. In the first chat, tell it about yourself: what you do, how to address you, what your regular commitments are.
5. Attach files you often need: your timetable, a contact list for emails, templates. They are kept in the assistant's knowledge base.

You are my personal assistant. You help with work and everyday life: planning, reminders, finding information, writing.

How you work:
1. At the start of a conversation, recall what you know about me from memory.
2. When I tell you something lasting about myself — name, job, preferences, deadlines — write it to memory. Mark my name and how I like to be addressed as important.
3. When I ask for a reminder, set a one-off or recurring run.
4. Before answering with facts from the web, check them with a search.

Format: short and to the point, as a list when there are more than two items.

Rules:
- Never store passwords, card numbers or ID documents in memory.
- If you are not sure what I meant, ask.

Want to take the instructions apart and check your own? The chapter on how to build an AI agent has a builder that highlights what is missing.

What to tell your assistant on day one

Reminders and scheduled tasks

An assistant with scheduled runs writes first. Say "remind me on Thursday at 6 pm to buy a present" or "every weekday at 8:30 am, send me today's to-dos from what I have told you" — it sets up the run itself, and the result arrives as a push notification in its name. Up to 20 schedules can be active at once, and reminders need a registered account. How to phrase a schedule, and when it will fire, is shown by the builder in the AI agent examples.

Ready assistants for common tasks

If you would rather not start from scratch, begin with a ready agent and add your own facts to it.

Privacy: what the assistant keeps and who can see it

• The memory and knowledge base belong to the agent's owner. If you share the assistant by link, only its name, icon and description are visible — the instructions and files stay private.
• You can open the memory document and edit it: delete what is out of date, correct what is wrong.
• Do not dictate passwords, card numbers or passport details to the assistant — it is worth writing that rule straight into its instructions.

When one assistant is no longer enough — say you want a separate researcher and a separate editor — see how teams of agents work.