Last updated

What Mira remembers

A generic chatbot forgets you the moment the tab closes. Mira doesn’t have to. As you work, it can keep a small set of long-term memories: durable facts and preferences it carries from one chat into the next, so you stop re-explaining yourself. And because this is Undra, those memories are not locked in a hidden database. They are ordinary notes in your workspace that you can read, edit, and clear.

What a memory is

A memory is a durable fact worth carrying forward, not the contents of a conversation. Things like:

  • how you like your summaries or release notes written
  • what your projects are called and who is on them
  • a standing preference (“always use British spelling”) or a hard rule (“never touch the vendor folder”)

Memory is always on and works quietly in the background. That is the difference from a skill: a skill is guidance you deliberately flag and Mira applies when it fits; a memory is something Mira picks up as you go, subject to your approval.

You are in control

Mira never silently decides what to keep. The flow is: Mira suggests, you decide.

Open the AI page and go to Memory (from Settings → AI → Open AI page, or the sidebar) to manage all of it in one place:

  • See everything. The Memory section lists every fact Mira currently holds about you. Nothing is hidden.
  • Approve or dismiss suggestions. After a conversation, Mira may propose a memory worth keeping. It waits in a Suggested list until you approve it or wave it off, so only what you accept sticks.
  • Pin the rules that must always hold. Pin a memory to make it an always-on constraint Mira carries into every chat, no matter how large your memory grows. Reserve this for real must-nevers (safety, allergies, “never do X”); pinning everything defeats the point.
  • Edit or clear. Fix a fact in place, delete one, or select several and clear them in a single confirm. Removal is permanent, so it asks first.

It scales without getting in the way

While your memory is small, Mira simply keeps the whole set in mind. As it grows, Mira brings forward the memories most relevant to what you are asking, plus anything you pinned, so a large memory stays fast and focused instead of drowning each reply.

Memories are just notes

Every memory lives as a real note in a Mira/Memory folder in your workspace, grouped by subject: one note per topic, each a short description of the topic followed by its facts as a plain list. There is no separate index file to keep in sync; the notes are the memory.

That means your assistant’s memory is as inspectable and portable as the rest of your work: open a topic note to read exactly what Mira knows, edit a bullet to correct it, or delete the note to forget the whole topic. It syncs, backs up, and travels with your workspace like any other file.

A managed folder, not a scratchpad

In the shipping app the Mira/Memory folder is a managed store, so you edit and curate the notes already there rather than hand-creating new ones inside it (a note dropped in by hand would be read as a memory). To teach Mira a fact, just tell it in chat, or add a bullet to the relevant topic note.

See it on the graph

Memory topics show up in Undraverse like any other note, marked with a small rose brain badge, and the “What Mira has learned” filter narrows the whole map to Mira’s memories and skills plus the notes they connect to. It is a quick way to see the shape of what your assistant knows and where it came from.

Where to go next

  • Mira: the assistant that builds and uses this memory.
  • Chats: where memories are formed and carried forward.
  • Setting up AI: connect a provider so Mira works at all.
  • Local-First Trust: why your data, memory included, stays yours.