Recall: see and change what an AI remembers about you
Part 4 of 4. A live prototype you can try.
Assistants remember things about you, but rarely show you what. Recall puts each remembered fact on a card that says where it came from and where it applies. You can fix it when it's wrong or delete it. It runs on synthetic data at a fictional company.
Live prototype: Recall. Data is synthetic.
The problem: you can't see what an AI remembers
Most assistants remember things about you but don't show you what.
Some facts you told it. Others it guessed. Some from one project turn up in another. You can't check or fix any of them. Recall doesn't make the assistant remember less. It shows you what it remembers.
Drag the handle. On the left, memory is one on/off switch. On the right, the same memory is shown fact by fact. Colors mark the state: indigo for In use, slate for Scoped to a project, coral for Withheld, rose for Forgotten.
See it
Every remembered fact is a card you can read.
Know where it came from
Each card says "you told me directly" or "inferred from 3 chats," and how sure the assistant is.
Choose where it applies
This project, everywhere, or off. A fact from work stays out of your personal chats.
Delete it
Delete one fact from one project or from everywhere. Nothing else goes with it.
Each remembered fact is a card
A card shows the fact, where it came from, where it applies, and a way to turn it off.
If the assistant guessed the fact, the card says how sure it is. You can edit, limit or delete each card on its own.
One memory card, running live in the Minia design system, like the app above.
Choosing where a memory applies
Where a memory applies is the main control.
A fact that helps at work may be unwelcome in a personal chat. Each memory can apply everywhere, to one project, or be switched off. Switch workspace and a panel shows what's in use and what's withheld.
What's used and what's withheld in each workspace, running live in the Minia design system.
Deleting one fact without wiping everything
"Clear all memory" deletes everything. Recall lets you delete one fact.
Before deleting, it names the fact in plain words and asks: remove it from this project, or everywhere? Nothing is deleted silently.
The delete step, running live in the Minia design system.
Try it in the app
Keep what helps, limit what fits only one project, and delete what you never wanted kept.
Keep or discard the four notes from the afternoon. Switch from the Strategy desk to the Job search workspace and see which memories drop out. Limit one fact to one workspace, bring one in from another, and delete the noisy guess. The memory log records every change.
How the design changed from v1 to v6
It started as one on/off switch. Each version added one more thing you could see or change.
By v6 you could check, fix, limit and delete each fact on its own.
-
1
A single "memory: on/off" switch
All or nothing. You could turn memory off, but not see anything it held.
-
2
+ A list of memories
You could see the facts, but not where they came from.
-
3
+ Source and confidence
Each fact marked "you told me" or "inferred," with how sure it was, so a guess no longer looked like something you said.
-
4
+ Edit in place
A wrong guess became a quick fix, not something to delete and teach again.
-
5
+ Where it applies
This project, everywhere, or off, with a live view of what's used where.
-
6
+ Exact delete · current
Deleting names the fact and where it goes from: one project or everywhere.
The other three prototypes in the lab
Steer agrees the brief before the AI writes. Ground shows where each claim comes from. Oversee sets what an agent may do alone.