How Your Personal Assistant Stays Yours
The Personal Assistant is the face FactoryOS shows you when it talks back. It's a chat surface, a daily-briefing sender, and a lookup helper for the people and decisions in your work, with a customizable visual character and an optional voice. It's also strictly yours — it sees your brain, your schedule, and your contacts, and never anyone else's. The longer you use it, the better it knows the kind of answers you want back.
How FactoryOS Talks to You
The assistant is the system's voice and personality, not a side feature. When FactoryOS sends you a morning briefing, prompts you about a meeting starting in fifteen minutes, or surfaces a follow-up from yesterday, the assistant is the messenger. The chat surface in the top bar is where you can ask it questions back — "what's on the schedule today," "what did we decide about the Q3 budget" — and the answer comes from the same persona, in the same tone.
You can use it as much or as little as you want. It works through the day whether you talk to it or not.
Scoped Only to You
The assistant only sees your second brain. Every person on a FactoryOS box has their own private brain — their own notes, their own meeting history, their own contact graph — and the assistant attached to a person is walled to that brain. It doesn't read across to anyone else's.
That isn't a settings checkbox; it's the architecture. Channels in FactoryOS partition data, and each user has a personal channel that no role and no override can reach into. The assistant inherits that boundary. If a colleague's assistant happens to know about a deal in progress, yours doesn't — unless you've separately been granted access to a shared channel that holds it.
Two Profiles That Evolve
The assistant carries two persona profiles, not one. The first is its own — the personality, the tone, the small mannerisms that make a particular character feel like a particular character. The second is a profile of you — what kind of answers you want, whether you prefer raw numbers or a digested recommendation, the topics you care about, the ones you brush off.
Both profiles update from feedback. Up- and down-votes on individual replies move them, and verbal cues during chat — "just give me the number, skip the explanation," "I always want the trade-offs spelled out" — move them faster. Over weeks and months, the assistant's tone settles into something that fits you, and its sense of what you want sharpens.
This kind of profile-building reads strangely when a cloud service does it — "the AI is keeping notes on me?" — but is no issue when it's local. The difference is who keeps the notes. On FactoryOS, the notes stay on the box. Your AI hardware, kept in your office. No third party stores them, no remote model trains on them, no other user on the same box sees them.
Your Daily Briefing
The assistant compiles a daily briefing from what your calendar, inbox, and brain know about your day. The format is a single message — meetings on the schedule, action items still open, follow-ups from yesterday, anything urgent that came in overnight — pulled together overnight or first thing in the morning, depending on how you want it delivered.
That briefing is also the assistant's deepest cross into the rest of the system. The mechanics of what it can pull from are covered in [what your morning briefing actually knows](what-your-morning-briefing-actually-knows), and the meeting-prep variant lives in [walking into meetings already prepared](walking-into-meetings-already-prepared).
A Light Touch Rolodex
The assistant doubles as a queryable directory. "What's Jordan's phone number," "when did we last meet with Acme," "who at our firm has talked to the new client," "summarize my last three calls with Marcus" — all questions the assistant can answer because the [brain](how-the-knowledge-graph-remembers-over-time) already has the people, the meetings, and the conversation history connected together.
It isn't a full CRM with pipelines and reporting. It's the lookup layer on top of the contacts and notes you already have — a light rolodex with memory.
A Character You Customize
The assistant has a visible character, and you build it. A short chat sets the base look — "a woman in her thirties, dark hair, business casual, warm but professional" — and once you render a few takes and lock the one you like, that becomes the assistant's identity from then on.
The locked look doesn't sit static. The system restages the same character to fit the moment: time-of-day cues in the image, seasonal details, holidays the calendar already knows about, the occasional change in setting. You see a morning version and an evening one, an autumn version and a winter one, without anyone touching a setting. Deeper customization — refining features, locking seeds, pinning scenes — lives in Persona Studio.
A Voice You Pick
The assistant can speak. Voice mode is a toggle on the chat surface, and when it's on, the assistant replies in a voice you've chosen from the available options. Several styles are supported — always-on listening, push-to-talk, or just having the assistant speak responses aloud — so the conversation feels however you want it to feel.
Voice is local to your box. Speech-to-text and text-to-speech both run on the same hardware your chat does, so nothing leaves the building — covered in more detail in [where your voice data actually goes](where-your-voice-data-actually-goes). Even if you never turn it on, the capability is built in and ready when you want it.