Designing Hathi for the whole family

Hathi was built for individuals. Extending it to a whole family meant rethinking memory, safety and shared context, here is how we approached it.

The Razormind team 3 min read
On this page
  1. The problem with a shared assistant
  2. Separate memory, shared household
  3. Safety that scales down
  4. What a family assistant should not do
  5. What stays the same
  6. An honest note

Hathi began as an assistant for one person: a companion in your messages that remembers you and helps you over time. That framing is simple to build and simple to trust. When we set out to make Hathi available to a whole family, we found that almost every design decision had to be revisited, because a household is not one user wearing several names.

The problem with a shared assistant

The obvious shortcut would have been to let everyone in a family talk to the same Hathi and pool what it remembers. We rejected that quickly. Memory built for one person becomes a liability when it is shared: a parent's work notes, a child's school worries and an adult's private plans should not settle into one undivided store. Shared memory is convenient until it is intrusive.

Separate memory, shared household

Our approach is to keep memory individual even when the family is shared. Each member of a family has their own Hathi memory, linked to them, not to a single household account. A parent can be confident that what they tell Hathi stays theirs, and a child gets an assistant that is genuinely oriented to them rather than to a blurred average of the household.

At the same time, a family is a group, and there are things a household does want to coordinate, plans, routines, shared reminders. The design keeps the two apart: individual memory for each person, and a small, explicit place for the shared things a family chooses to keep together.

Safety that scales down

Designing for children changed the tone as well as the structure. A family assistant has to be appropriate for the youngest person using it, which means reconsidering how Hathi speaks, what it is willing to discuss, and how it responds when a conversation moves somewhere it should not. We treat those guardrails as part of the product, in the same spirit as Jen, our family image generator.

The guiding rule is that a parent should be able to understand what a family Hathi will and will not do, and to feel comfortable with the answer before a child starts using it.

What a family assistant should not do

A good deal of family design is subtraction. A family Hathi should not quietly combine one person's words with another's, should not treat a child's casual remark as a lasting fact without care, and should not present itself as an authority on matters a parent ought to decide. These are not edge cases; they are the ordinary situations a household creates, and each one shaped a decision we made.

Being explicit about those boundaries is what lets a family extend trust. Parents can see the shape of the system before a child meets it, which is the only reasonable place to start.

What stays the same

For all the changes, the foundations are the ones Hathi was built on. Memory is still explicit and still yours to inspect and delete, a child's memory and a parent's alike. The assistant still runs in the messages you already send, so there is nothing new for a family to learn. And Hathi still names itself rather than hiding behind a generic assistive voice, because a family should know who they are talking to.

An honest note

A family is not a monolith, and we would be wary of any single design that claims to suit every household. What we have aimed for is a structure that keeps individual memory intact and lets shared context be chosen deliberately rather than assumed. Where families differ from that model, we would rather hear about it than pretend the fit is perfect.

Hathi is live at hathi.chat. If you would like to bring it into a household, the memory and safety controls are there to look over first.

Razormind

Systems that reason, remember and act

Hathi, Jen and Consul are built on one shared foundation. If you would like to talk about where that fits, we would be glad to hear from you.

Related reading

The architecture Razormind is built on

Razormind systems are designed from first principles: explicit structure, verifiable reasoning and persistent memory. Here is what that means, and how the three products share it.

3 min read

← All posts