All use cases

Memory by use case

AI memory for personal assistants

An assistant that knows your week, your people and your preferences.

Why memory matters here

A personal assistant is only personal if it knows the person. The useful facts are spread across months of conversation: a dietary need mentioned once, a manager's name, the fact that mornings are kept free. No context window holds all of that, and nobody wants to repeat it.

Who a memory belongs to

One memory per person, private to them, which they can read, correct and delete.

The moment it shows

Asked to book dinner, it picks a place that suits a diet you mentioned in spring.

What to remember

  • Standing preferences: food, travel, working hours, tone.
  • People and how they relate to the user.
  • Ongoing projects and their next step.

What to forget

  • One-off plans once the date has passed.
  • Anything the user asks it to forget, at once and completely.
  • Guesses. Store what the user said, not what the model inferred about them.

See recall before you build it

Add a few facts from your own use case and ask a question.

Personal assistants: common questions

How much should a personal assistant remember?

Less than you might expect. A few hundred short, well-chosen facts serve better than a full transcript, because the right one is easier to find.

Should the user be able to see what is remembered?

Yes. A memory the user cannot read, edit or delete is hard to trust and may fall short of privacy law in many places.

Other use cases

Pack the doko. Ask it anything.

Nine memories, one question, no account. See which facts an agent would carry into its next answer.