Companion Memory personas
The people on this page are fictional. They are invented to show who Companion Memory is built for. They are not real users, and nothing here is a testimonial.
Each section describes one person. It covers who they are, what they want from memory and sharing, what Companion Memory does for them today, what is not yet built, and what Companion learns for them at work and at home. How Companion learns explains those learning features. Each section ends with the scenarios where the person appears. The scenario pages come later, so this page lists only their titles.
Four status labels appear on this page. They mean the same thing in every section:
- Works today: the capability is on by default.
- Behind a setting: the capability is in the code, but it is off until an owner or operator turns it on.
- Not yet: the capability is not built, or it is built but not connected to the product.
- Planned: the design is documented, and the code does not exist yet.
Companion Memory is the personal memory product. It holds your messages, photos, calendars, documents, and the people in them. You can search that memory, ask other people’s Companions questions, and share carefully from it.
This diagram groups the personas by what they mostly rely on Companion for: family and home (Maya and Ruth), friends and community (Devon, Ana, Tomas, and Priya), work and clients (Sam Okafor and Noor), and research and consent (Dr. Amara Osei).
flowchart LR P["The nine personas"] --> F["Family and home<br/>Maya, Ruth"] P --> C["Friends and community<br/>Devon, Ana, Tomas, Priya"] P --> W["Work and clients<br/>Sam Okafor, Noor"] P --> R["Research and consent<br/>Dr. Amara Osei"]
Maya
Maya is 38. She is the household operator for her family. She bought a family box for photo memory and privacy. She is a capable user, but not a developer. When something in the household needs sorting out, she is the person others ask.
What she wants. Maya wants to ask family members for photos and documents without chasing them. She wants standing permissions for routine questions. She wants to see what she has shared and what she has declined.
How Companion helps today.
- Works today: standing grants that answer routine asks, and revoking a grant she no longer wants.
- Works today: Disclosure History, which lists what was shared and what was declined.
- Behind a setting: web push reminders in a browser. Push needs keys set on the box.
What is not yet built.
- Not yet: asking a peer to remove copies it already holds after a grant is revoked.
- Not yet: push notifications in the phone apps.
- Planned: a care-circle digest for an older relative. It would show no location and would end on a set date.
What Companion learns for Maya.
- Works today: suggested to-dos from her messages, such as a request from a relative. On the default Balanced setting, the ones Companion is confident about go straight onto her to-do list. The rest wait for her to accept or dismiss them.
- Works today: the Calendar view colors the next seven days with what she usually does at those times. Companion learns this from her own timeline on the family box, and nothing is sent to a cloud service to be learned from.
- Planned: spotting promises in family conversations, such as “I’ll send the photos on Sunday”, and staging a task or calendar event for her to confirm. She could then follow up without chasing anyone.
- Planned: insights that interrupt her only when they are worth it. For the person her household turns to, that would mean fewer alerts.
Scenarios. Share an album with grandparents. Keep a care circle informed.
Devon
Devon is 29. He is an early adopter who runs his own box on a network-attached storage device. His three closest friends run boxes too. He reads technical specifications closely, and he wants the network of boxes to belong to the people who run them.
What he wants. Devon wants to ask his friends’ boxes questions under rules he sets. He wants to see who can ask him what. He wants to reach his storage device from his phone when he is away from home, and to recover quickly if a device is lost.
How Companion helps today.
- Behind a setting: sending an ask to a friend’s box, and answering asks that friends send you. Both boxes need their peer settings, and the two owners need a peer link.
- Works today: standing grants for each peer.
- Behind a setting: opening his box over Tailscale in a phone browser, once Tailscale is set up on the box and on his phone.
- Works today: revoking a grant for a friend. New asks from that friend stop.
What is not yet built.
- Not yet: pairing with a friend by scanning a QR code.
- Not yet: asking a peer to remove copies it already holds.
- Not yet: getting a friend’s answer back to your box.
- Not yet: verifiable credentials that prove who a person is without a shared secret.
- Not yet: decentralized identifiers for peer identity.
What Companion learns for Devon.
- Works today: each night his box updates a small personal model of his routine from the day before. An update that fails is not saved, and since Companion Memory 2026.11.11 earlier versions are kept so a bad one can be rolled back.
- Works today: a quick-decision service that apps and agents on his box can call. It answers yes-or-no and choice questions by comparing meaning. It is on by default, and he can turn it off under Settings → Model. No built-in feature uses it yet.
- Behind a setting: the Forecast page, which shows how his next day and week are likely to go. As the operator of his own box, he can turn it on.
- Not yet: running training on another machine he owns, such as a desktop with a bigger graphics card, through Companion Hub.
Scenarios. Use the Hub away from home. Recover after losing a device. Check a claimed identity before you answer.
Ruth
Ruth is 71. She is a grandmother and the family archivist. She keeps about fifty years of photos, letters, and recipes. She worries about other people seeing her things. She uses a voice assistant and a tablet.
What she wants. Ruth wants each pending request explained in plain words before she decides. She wants to keep her archive private from strangers. When she has a box of her own, she wants to move the family circle onto it in one step.
How Companion helps today.
- Works today: revoking a grant she no longer wants.
- Works today: shared search, which is on by default. Anyone signed in with the link can read it. Each read by another person is recorded, and you can revoke one reader.
- Works today: export and import of your memory data, which moves the archive to a new box. Import replaces the data already on the receiving box.
What is not yet built.
- Not yet: a preview that shows what an import will replace before it runs.
- Not yet: a check on the receiving box that an export is complete and unchanged before the import starts.
- Not yet: moving sharing settings with the archive. Grants, peer links, household tiers and asks are not in the export.
What Companion learns for Ruth.
- Works today: Companion learns her usual week from her own timeline on the family box, and shows it in the Calendar view. Nothing is sent to a cloud service to be learned from.
- Works today: she can choose the Cautious setting in Settings → Autonomy. With it, every to-do that Companion suggests waits for her to accept or dismiss it.
- Not yet: a small model, trained on her own choices, that would answer quick questions such as whether a new photo or letter is worth remembering.
- Planned: smarter search ranking, so the recipe or letter she means would come up first in fifty years of material.
Scenarios. Share an album with grandparents. Move the archive to a new box.
Sam Okafor
Sam Okafor is 45. He works as an event photographer. Attendees ask him whether he has photos of them. He answers from his archive, and he sends only what each person is entitled to receive.
What he wants. Sam wants to answer “do you have photos of me?” without handing over the whole archive. He wants to review every answer before it leaves his box. He wants to charge for some deliveries.
How Companion helps today.
- Works today: each request waits in his review inbox, and he sees each answer before it is sent.
- Works today: he can answer with less than was asked, such as a yes or no, or a summary.
- Works today: he sets the trust tier for each person who asks him.
What is not yet built.
- Not yet: a rule for attendee photo deliveries. Photo delivery to a stranger needs a rule that the product does not have yet.
- Not yet: priced requests and payment for a delivery.
- Not yet: verifiable credentials that show an asker is who they say they are.
What Companion learns for Sam.
- Works today: suggested to-dos from his messages, such as sending a gallery to an event organizer. The ones Companion is confident about go straight onto his list, and the rest wait for him to accept or dismiss them.
- Not yet: small models, trained on his own past choices and run on his box, that would answer quick questions about a request, such as whether it needs his attention today.
- Planned: a personal tone add-on that learns his writing style, so draft replies to attendees sound like him. He would still see each answer before it is sent.
Scenarios. Check a claimed identity before you answer.
Dr. Amara Osei
Dr. Amara Osei is 52. She is a sleep researcher. She recruits volunteers for a study. She wants each volunteer to see what is asked, why, and what they are paid. She also needs consent records that she can show to an ethics board.
What she wants. Amara wants to send one research request to many volunteers, with a scope, a price, and a purpose. She wants a consent record for each disclosure. She wants aggregate results that contain no individual rows.
How Companion helps today.
- Works today: each volunteer’s box decides its own answer, under the rules that person set.
- Works today: answers can be a yes or no, or a summary. The tier and the question decide which one a person gets.
What is not yet built.
- Not yet: priced requests, and a payment path for volunteers.
- Not yet: one request that reaches people on other boxes. Group requests reach only people on the same box.
- Not yet: a reason shown to the asker when a request is declined. A declined request looks the same as one that is still waiting.
- Not yet: aggregate questions across many volunteers.
- Planned: consent receipts that show the consent basis for each disclosure.
What Companion learns for Amara.
- Works today: each volunteer’s box sorts that person’s days into broad activities, such as work, exercise, and sleep, and learns their usual week. That learning stays on the volunteer’s box. Asking about those patterns across many volunteers needs aggregate questions, which are not yet built.
- Works today: suggested to-dos from her messages, such as a volunteer’s question that needs an answer. The ones Companion is confident about go straight onto her list, and the rest wait for her to accept or dismiss them.
- Planned: quality checks that every newly trained model must pass before Companion uses it. They will test accuracy, and whether a model’s confidence matches how often it is right.
Scenarios. Check a claimed identity before you answer.
Priya
Priya is 33. She is new in town and is building a social circle. She is careful about who she trusts, especially before she meets someone in person. She uses Companion Memory as a stranger on other people’s boxes, and she uses it to check a few claims.
What she wants. Priya wants to confirm that someone she met at a conference is who they say they are, without sharing her own details. Before a first date, she wants to check one mutual contact and an age range, and nothing more.
How Companion helps today.
- Works today: she decides each request before anything is sent.
What is not yet built.
- Not yet: confirming a claim such as “we met at the conference” with a verifiable credential.
- Not yet: a safety check that shows an age bracket and one mutual contact.
- Not yet: checks on mutual contacts.
- Not yet: community requests, where a club posts one request for its members across many boxes.
What Companion learns for Priya.
- Works today: Companion learns her usual week from her own timeline and shows it in the Calendar view. Social time is one of the activities it sorts her days into. In a new town, the pattern starts thin and fills in as her weeks repeat.
- Planned: spotting plans she agrees to in conversations, such as coffee with someone she met at a conference, and staging a calendar event for her to confirm.
- Planned: spotting duplicate contacts, so someone she met at a conference and later messaged would show up as one person.
Scenarios. Check a claimed identity before you answer.
Ana
Ana is one of Devon’s peers. She has her own box, and she receives his requests. She reviews each request before she answers it, and she can answer with less than was asked.
What she wants. Ana wants to review every request before anything leaves her box. She wants to stop answering a friend who leaves her circle. She wants a clear way to know that a request really comes from the person it names.
How Companion helps today.
- Works today: she reviews each request before anything leaves her box, and she can answer with less than was asked.
- Works today: revoking a grant stops new requests from that person.
- Works today: Disclosure History lists what she has shared.
What is not yet built.
- Not yet: a credential that proves who a person is without a shared secret.
What Companion learns for Ana.
- Works today: Companion learns her routine on her own box. Each night it updates a small personal model from the day before, and nothing is sent to a cloud service to be learned from.
- Works today: she can choose the Cautious setting in Settings → Autonomy. With it, every to-do that Companion suggests waits for her to accept or dismiss it, the same way every request does.
Scenarios. A member leaves a group. Check a claimed identity before you answer.
Tomas
Tomas is in his thirties or forties. He organizes a standing group of five friends from three households. He plans dinners and films for the group, and he buys tickets when the group agrees.
What he wants. Tomas wants to agree a dinner or a film with the group without a long thread of messages. He wants to see when each friend is free, without seeing their event details. He wants the group’s membership to stay clear, and he wants to buy tickets inside a limit that the group agrees.
How Companion helps today.
- Works today: plan containers for a group plan.
- Behind a setting: a free-or-busy check with each friend’s box. The answer shows free or busy only, with no event details.
- Works today: on the default Balanced setting, anything that sends or changes something outside the box waits for approval, and so does an outside lookup made after Companion has read his memory in the same reply. An outside lookup he asks for directly runs at once.
- Behind a setting: web push reminders in a browser. Push needs keys set on the box.
What is not yet built.
- Not yet: one availability answer for the whole group.
- Not yet: RSVP tracking. Replies to a plan are not recorded.
- Not yet: buying tickets with an agent, and a spending limit for purchases.
- Not yet: removing a former member’s copies of a plan.
- Not yet: reminders inside the app.
What Companion learns for Tomas.
- Works today: suggested to-dos from his messages, such as booking a table once the group agrees. On the default Balanced setting, the ones Companion is confident about go straight onto his list. The rest wait for him to accept or dismiss them.
- Works today: the Calendar view shows his usual week, which helps him pick an evening to suggest. It covers his own routine only. Each friend’s routine is learned on that friend’s box and stays there.
- Planned: spotting promises in group conversations, such as “I’ll pick the restaurant”, and staging a task or calendar event for him to confirm.
Scenarios. Plan a dinner or film with a group. A member leaves a group. Buy tickets inside a spending cap.
Noor
Noor is in her thirties or forties. She is an independent consultant who runs client meetings. She prepares agendas, invites clients, keeps minutes, and decides how long each client’s material should stay.
What she wants. Noor wants each client’s material to stay separate from every other client’s. She wants to send a calendar invitation that a client can reply to. She wants minutes that she can trace back to their source, and she wants a retention rule for each matter.
How Companion helps today.
- Works today: plans with an agenda, an attendee list, and notes in one place.
- Works today: calendar import from Outlook or Google. The Google import reads the primary calendar only.
- Works today: calendar invitations as a file to download. The file does not record replies, so a client’s answer does not come back.
- Works today: forget a person or an item. You can archive it or delete it permanently. Deletion applies on your own box only.
What is not yet built.
- Not yet: client replies to invitations.
- Not yet: separate matters. Search and grants apply to the whole account, not to one matter.
- Not yet: retention rules for each matter or each client.
- Not yet: removing a client’s copies from other boxes.
What Companion learns for Noor.
- Works today: suggested to-dos from client messages, such as sending minutes after a meeting. The ones Companion is confident about go straight onto her list, and the rest wait for her to accept or dismiss them. Because matters are not yet separate, the suggestions draw on all her clients at once.
- Planned: a personal tone add-on that learns her writing style, so drafts of client emails sound like her.
- Planned: spotting commitments in meetings and emails, such as “we will send the revised plan on Monday”, and staging a task or calendar event for her to confirm.
Scenarios. Prepare a client meeting. Use the Hub away from home. Check a claimed identity before you answer.
Scenario titles
These are the scenario titles used on this page. The scenario pages come later.
- Plan a dinner or film with a group
- Prepare a client meeting
- Share an album with grandparents
- A member leaves a group
- Keep a care circle informed
- Use the Hub away from home
- Recover after losing a device
- Buy tickets inside a spending cap
- Check a claimed identity before you answer
- Move the archive to a new box
Related pages
- How Companion learns: the learning features on this page, in plain language.
- Companion at work and at home: everyday examples of what Companion learns.
- Personal AI on your own hardware: what is planned for personal models.
- AI training, inference & jobs: the technical details.