How Hathi remembers, and how you delete a memory

Hathi is built to remember you over time. Here is what that memory is, how Hathi draws on it, and how you review and delete it at any time.

The Razormind team 3 min read
On this page
  1. Memory as a component
  2. What Hathi remembers
  3. How recall works
  4. Memory that follows you
  5. Reviewing and deleting
  6. An honest limitation

Most assistants forget you the moment a conversation ends. Hathi is built on the opposite premise: that a personal assistant should remember you over time, your preferences, the people and projects you mention, the way you like to be spoken to. That memory is useful only if it is also trustworthy, which is why Hathi treats memory as a component you can see and manage rather than something hidden.

Memory as a component

Hathi runs inside the messaging apps you already use. As you talk, Hathi keeps a store of durable facts about you, an explicit collection that can be read, corrected and removed. This is a deliberate design choice. When memory is an explicit component, the assistant can be honest about what it knows, and you can hold it to account.

It is worth being precise about scope. Hathi's memory is tied to you as a person, not to the device you happen to be holding. Support for family members is described in a separate post; here we are concerned with the individual.

What Hathi remembers

Hathi aims to remember the things that make it more useful and less repetitive:

  • Your preferences, how you like replies framed, the tone you prefer, the level of detail you find helpful.
  • People and projects you mention, so that Hathi can refer to them correctly later.
  • Recurring commitments and routines you have asked Hathi to help with.

Hathi does not try to remember everything. A memory that hoards every passing sentence is noisier and harder to keep accurate. The goal is a small, dependable store of facts that genuinely improve the next conversation.

How recall works

When you send a message, Hathi looks at what you have said before and brings the relevant memories into the reply. Because memory is explicit, Hathi can also tell you which of your saved facts it used when you ask. Recall is anchored to you: a memory saved in your conversations is used for your conversations.

Memory that follows you

Because Hathi's memory belongs to you rather than to a single thread, it carries across your conversations. A preference you set today does not have to be restated tomorrow, and a fact you correct once stays corrected. This is the practical difference between an assistant that remembers and one that merely has a long context: a long context forgets as soon as it overflows, while a store of explicit facts can persist and improve over time.

That persistence is also why we keep the store small and deliberate. Every fact that survives is one Hathi can be questioned about later, so the value of the memory comes from its accuracy rather than its volume.

Reviewing and deleting

You are in control of what Hathi keeps. You can ask Hathi what it remembers about you, ask it to correct a fact that has changed, or ask it to forget something. When you delete a memory, Hathi removes it from your store. A request not to use that fact again is honoured by deleting it.

Deletion is designed to be straightforward and to need no special ceremony. If you want to remove everything, you can ask for that directly. If a single fact has become wrong, an old address, a project that has finished, you can remove that on its own and keep the rest.

An honest limitation

Hathi's memory is only as good as what it has been told. It can mishear, save something you did not mean to keep, or hold a fact that has since changed. Precision and correction are exactly why the store is visible: when Hathi gets something wrong, the fix is to look at what it holds and adjust it. We would rather give you a small, inspectable memory than an impressive, opaque one.

Hathi is live today at hathi.chat. If you would like to try it, the memory store and the deletion controls are there from the first conversation.

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

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.

3 min read

← All posts