Documentation
How to Use Mirror’s Memory Controls
A practical guide to how Mirror remembers, how users stay in control, and how memory can be viewed, exported, or deleted.
Mirror is designed to become more personal over time. It can remember themes, names, patterns, and preferences that help future conversations feel more relevant and supportive.
But memory must always belong to the user.
Mirror’s memory is not designed to secretly collect information or create a hidden profile. Its purpose is to support continuity. If a person returns to the same theme many times, Mirror can gently reflect that pattern. If someone shares what helps them feel grounded, Mirror can remember that too.
For example, Mirror may remember that walking helps after a stressful workday, or that Sunday evenings often feel heavy. It can later use that memory carefully, as an invitation, not as a label.
A good memory response might sound like this:
“You mentioned before that walking helps when your thoughts feel crowded. Would that feel useful tonight?”
A bad memory response would sound too fixed or judgmental:
“You always get anxious on Sunday nights.”
Mirror should avoid labels. Memory should help the user feel seen, not watched.
Users should be able to open Memory Controls and see what Mirror remembers. Each memory item should have a clear title, category, date, and available actions. The user should be able to edit, export, or delete it.
Memory Controls should include four basic actions:
View memory — see what Mirror has saved.
Edit memory — correct anything that feels inaccurate.
Export memory — download a copy for personal use or backup.
Delete memory — remove one memory or reset all saved memory.
This is important because emotional AI depends on trust. A user should never feel trapped by what the system remembers. They should always be able to reset the experience.
Mirror should also avoid saving certain information automatically. Highly sensitive health details, crisis-related statements, financial information, legal information, and private information about other people should not become ordinary memory without clear reason.
If a conversation includes crisis, harm, or clinical depth, Mirror should not treat that as normal memory. It should respond carefully and point the person toward real-world human support.
Mirror is not a therapist, doctor, or emergency service. Memory should support reflection, not replace care.
The recommended beta version of Mirror memory should include:
memory enabled by default only after clear consent;
memory review available at any time;
sensitive memory saved only with confirmation;
export and delete controls visible;
crisis-related content excluded from normal memory;
local-first storage where technically possible.
The rule is simple:
Your memory belongs to you. Mirror only holds it while you allow it to.
If you want to help test Mirror’s first memory experience, join the waitlist at mirror-mirror.space.
Mirror, Mirror
Want to help shape Mirror?
Join the waitlist and become one of the first people to test a voice-first AI presence for the home.