Staff
Your Staff is a personalized digital operator that works continuously across your work and life. It maintains context, prepares what matters before you ask, challenges your thinking and helps turn decisions into action. You remain in control while your Staff builds knowledge, tracks commitments and keeps approved work moving between interactions.
Built around you
Your decisions do not fit neatly into separate systems. Work, family, travel, relationships, interests and financial responsibilities compete for the same time and attention. Your Staff starts with that reality: it brings together the context needed for the decision in front of you while keeping the boundaries between sources, roles and responsibilities intact.
A team that knows your context, supports your goals, and works with the people and tools already in your life.
Talk to your Staff, wherever you work
StaffMy Staff
Business Operator
Mike
Your Staff workspace
Connected, integrated understanding across all aspects of your work, your ambitions, and your personal and family agenda.
Works with your everyday tools
Connects your business systems
Draws on a choice of AI models
Connected through existing integrations, MCP servers and APIs. Which of these are in place is decided per deployment; the marks above name what Staff is built to work with, not what is switched on for you.
Shared understanding. Respected boundaries. You stay in control.
What it looks like when a decision reaches you
Decision board · Since yesterday
Handled
Four dependent items re-sequenced. Jack and Amit’s approvals remain on hold.
Kate confirmed at 07:20 that Thursday’s signing has moved to Monday. The four dependent items now follow that date. Jack and Amit’s approvals remain on hold until signing is confirmed; no new approval is needed from you to maintain the hold.
Needs your decision
Marketing is holding Thursday’s release.
RecommendedWait for the signed copy.
The release was planned around Thursday’s signing. With signing now expected on Monday, announcing on Thursday would mean announcing before there is a signed copy. That affects Insprod and the board pack.
Before deciding, confirm the revised signing expectation with Kate and check whether Marketing has made any external commitment to Thursday.
Company · Staff at work
An email moves the signing date from Thursday to Monday.
The agreement, your calendar, the board pack and the announcement plan.
A revised plan and one decision: should the announcement move too?
Commitments, commercial context, board material and internal decisions. Usable for the work in front of you, and not silently reusable everywhere else.
Each domain keeps its own identity, permissions and source and trust boundaries. Staff works across all of them as one operator rather than a collection of separate assistants — using what is relevant and permitted for the decision in front of you, without merging the domains into one pool. The set is extensible; these are the ones Staff operates across today. .
GUIDE is the set of principles behind that governance, applied the same way in every domain: , , , and .
One continuous day
Luke leads the commercial business for interactive retail displays at a European specialist monitor company. He owns revenue, product direction and the P&L while the business expands into the United States. His day crosses customers, product, people, budget and delivery — alongside lunch, an airport pickup and a late call from home.
Your Staff is helping you even while you are doing something else. In Luke’s day, Laura, Sam and Amara prepare before he asks and keep agreed work moving. He can work with them directly; his Staff brings their work together when it spans more than one part of his day.
Nine moments. One continuous operating context.
Luke is brushing his teeth. He hasn’t opened his calendar. His Staff is already fitting the moved commercial review, Sue’s 16:40 arrival and two overnight carry-overs into his day.
Morning brief
Laura
The review is now at 09:15. Commercial has the revised time; I’ve moved the preparation ahead of it.
Sam
The airport journey is already in the plan. The late call still works from home.
Staff
I’ll bring the work and personal updates together for breakfast.
Luke is at his desk, opening the brief for his first meeting. The commercial review starts in fifteen minutes. Pricing and a product dependency need Luke’s judgment.
Commercial review
Laura
The review has two decisions: pricing and the product dependency. The rest is an update.
Amara
The latest numbers are attached. The lower price only works at the proposed margin if the delivery assumption holds.
Laura
Then I’d hold the price for now. Luke, are you comfortable taking that position into the review?
Luke is working on the US expansion plan. Yesterday’s post has 61 reactions and 18 substantive comments. Kate, MD at SouTech, is among them; SouTech is already in the pipeline.
Reaction summary
Laura
Kate’s comment is part of a pattern: several people are asking about continuity during rollout.
Amara
SouTech’s account notes raise the same concern. That supports a follow-up; it doesn’t establish a buying decision.
Laura
I’ve drafted a reply about who looks after the rollout, plus a note for Pete. I’ve kept pricing out of it. Shall I send them?
Luke is reviewing the proposal and asks Amara how ready the research is. The evidence has continued to develop since their last conversation.
Research position
Luke
How ready is the research for the proposal?
Amara
The financial review is current and the Bloomberg forecast still fits the evidence. I wouldn’t use the roadmap claim yet; I haven’t finished testing it.
Luke
Leave that claim out for now. Keep going with the comparison.
Luke is leaving the restaurant after lunch. Sophie’s birthday is Saturday; the flowers ordered on Tuesday are ready two streets away.
On your route
Sam
Sophie’s flowers are ready at Bergmann’s, 200 metres along the walk back. There’s time before the next commitment.
Laura
The afternoon preparation is ready. Nothing at work needs that window.
Sam
You can collect them on your way back. They’re held until Friday if you’d rather leave it.
Luke is focused on the proposal, while Laura follows the release discussion. Marketing and Product have spent two days debating the release date. Across the messages, Laura sees a different constraint.
Release timing
Laura
The thread says release date. Three separate messages point to the same missing allocation of people.
Amara
The resourcing plan confirms the gap. The budget contains a possible offset, but Finance needs to validate it.
Laura
I’m having Finance check that offset. If it holds, would you rather use it to protect the date or accept a delay?
Luke is packing up his desk before leaving for the airport. The next meeting’s brief is unread, and the research comparison is ready to discuss on the drive.
Continue by voice
Sam
Leave at 15:45 for Sue’s 16:40 arrival. The journey has been reserved all day; the timing still works.
Laura
The brief is down to two decisions. Nothing requires Luke to stay at his desk.
Staff
We can carry on in the car. Would you like the meeting brief first, or the roadmap comparison?
Luke is driving home after the airport pickup, talking with his Staff. Dinner at 19:00 is now tight; tomorrow’s first meeting has shifted later.
On the way home
Luke
Move dinner an hour later.
Sam
The restaurant has confirmed 20:00.
Sam
Your first meeting tomorrow has moved to ten. Would you like your alarm thirty minutes later too?
Luke is wrapping up for the day before spending the evening with Sue. The agreed proposal comparison continues in the background; today’s feedback shapes tomorrow’s briefing.
Background work
Amara
I’m continuing the agreed roadmap comparison. I’ll check contradictions and return with the next useful question.
Oz
Tomorrow’s briefing can lead with that question and keep the supporting detail underneath.
Luke
Make that the default: decision first, detail underneath. And the afternoon update appeared twice.
1–3 of 9 moments
Built for trust
Staff is designed to become more capable without becoming less accountable. It can prepare broadly, reason across several sources and keep bounded work moving. What it may know, decide and do remains governed by identity, source, policy, risk and your authority.
The Personal Operator Architecture is the operating foundation behind Staff. It maintains current state, knowledge, attention, reasoning, authority, execution and evidence as Staff assets — independent of any single model, tool or interface. We call the discipline .
That allows new models, devices and integrations to strengthen Staff without taking ownership of your context or changing who remains in control.
Explore the responsibilities that keep context, capability and authority working together.
Layer 01
Operational Visibility — What the operator can see, steer and be answerable for.
Layer 01
Business and private sources retain separate identity, memory and policy boundaries. The operator receives unified briefings; models and workloads can run locally or in the cloud within those boundaries.
Download the one-page architecture overview →One operator view · Governed by design · Private by default · Auditable end-to-end · Enterprise-grade control · Multi-model & hybrid · Continuously learning
The operator race
Within seven weeks, xAI, Meta, Apple, Microsoft and OpenAI each introduced a personal AI that learns your context and acts across your apps. The contest is no longer which model is smartest this month. It is who becomes your digital operator, and Staff is built to be one that belongs to you, not to a platform.
xAI introduced Grok Bot, Meta introduced Muse, Apple began rolling out the Siri AI assistant in beta, Microsoft brought Autopilot agents in private preview into the new Copilot and OpenAI introduced Dots. A digital operator is the system that knows your context, remembers your preferences, acts across applications and asks for permission when the stakes are real. Whoever holds that position holds the layer between intent and execution.
Each wave of computing had a winner that owned the layer between people and what they wanted.
Google · Search
89.94%
of worldwide search engine use, September 2026
Google won search: the layer between a question and the web.
StatCounter Global Stats ↗Apple · Handheld computer
65%
of global premium smartphone sales ($600+ wholesale), H1 2026
Apple won the handheld computer, and made identity part of the product.
Counterpoint Research, 5 Aug 2026 ↗Amazon · Online commerce
40.5%
of US retail e-commerce sales, eMarketer forecast for 2025
Amazon leads online commerce: the layer between a shopper and what they buy.
eMarketer forecast, 2025 ↗The next winner may own the operator layer: the one between your intent and its execution.
Each new operator arrives through a channel its company already owns. The Siri AI assistant is built into iPhone. Muse works inside WhatsApp. Grok Bot can be tagged on X where available. Dots live in ChatGPT. Copilot is the enterprise version of the same play: the operator bundled into your employer’s Microsoft 365. Distribution is their advantage, and for users it is a lock-in. Once an operator holds your context, permissions and history, changing it feels less like switching apps and more like replacing someone who knows your life.
So the open question is loyalty. Will people stay with the operator bundled into the channel they happen to use, or follow the one that knows them across every channel? And when you change jobs, an operator that lives in your employer’s tenant stays with the employer.
That is why we built Staff the way we did: not another chatbot, but an operating layer for decisions, work and accountability that belongs to you rather than to a platform. A team of named specialists across your job, your businesses and your family, using the models that suit each task and working from knowledge you govern. The same governed operator, with your memory, context and permissions, follows you across Telegram, the web workspace and Control, and is designed to reach more channels. Its Chief of Staff is built to stay with you from one employer to the next, keeping each employer’s confidential material isolated while what it learns about how you work carries across.
Compare the approach
Pick an operator to see how its published approach compares with Staff on seven dimensions. Their columns summarise each company’s own announcements; Staff’s describe how Staff is designed to work.
BdG AdvisoryStaffCommon groundLike Staff, Grok Bot is persistent, gives its agents names and jobs, keeps working between conversations and comes back when it needs your approval.
StaffGrok Bot organises agents around tasks on one shared computer. Staff organises a governed team around your life, with boundaries that hold between its parts.
| Dimension | Staff | Grok Bot | Dots | Muse | Siri AI | CopilotEnterprise |
|---|---|---|---|---|---|---|
| Who works for you | A standing team of named specialists built around your life: a Chief of Staff for your job, a Business Operator, a Family Steward, a Researcher and more, each with a defined role and coordinated by your primary operator. | In beta: named Bots you create, each with one primary job such as inbox, expenses or recruiting; they can coordinate with each other in group chats. | Your dot: one always-on agent, rolling out gradually to eligible plans, with the option to add more dots announced for the future. | One personal agent, Muse, rolling out in the US, that turns your goals into a plan and advances the work on its own. | One assistant: a completely reimagined version of Siri, built on Apple Intelligence and rolling out in beta. | Copilot, plus Autopilots in private preview: always-on agents with their own identity, such as Scout, which works across your Microsoft 365 apps. |
| Models | Several providers. Models are chosen per task by capability and risk, higher-risk work can be reviewed by a second model, and any model can be replaced without losing your context. | An xAI agent platform. xAI’s launch materials do not describe a choice of model provider. | Powered by GPT-6 Astra, OpenAI’s own model. | Powered by Muse Spark, Meta’s own model, built for agentic work. | Apple Foundation Models, custom-built in collaboration with Google and its Gemini models, on device and in Private Cloud Compute. | Auto routing weighs accuracy, speed and cost to send each request to the model best suited for the job. |
| Memory and knowledge | Maintained knowledge with its sources and provenance, held as your asset outside any single model, so it survives every change of model, device or interface. | Each Bot keeps its own memory of working preferences, important facts and summaries of its prior work. | Draws on your ChatGPT memories and creates its own, including from connected apps. | Remembers what matters to you, learns from conversations, and forgets what you ask it to. | Personal context understanding across your messages, emails, photos and more. Siri Recap, summaries of conversations to review later, is announced for beta in late 2026 and requires Apple Watch Series 12 or Ultra 4 with a supported iPhone. | Builds context through Work IQ, learning how you work; an Autopilot (in preview) has its own memory inside your organisation’s tenant. |
| Authority and accountability | Authority governed by identity, source, policy and risk. Consequential actions come to you first, autonomy is earned domain by domain, and actions leave an auditable record. | Approval prompts for sensitive actions (allow once, always allow, or deny), with saved rules and admin policies for teams. | Asks for approval before sensitive actions; for the most sensitive, such as moving money, you take over yourself. | Checks with you before sensitive actions such as sending an email or making a purchase; you choose each app’s access. | Carries out systemwide app actions you ask for, such as drafting an email or editing and sharing a set of photos. | For its preview agents, sensitive actions can require a human sign-off, with a governed Entra identity, Purview data protection, audit, and admin controls in Agent 365. |
| Channels and continuity | Each specialist keeps one conversation that continues across Telegram, the Staff web workspace and Staff Control. Reply where you are; the context follows. | Desktop and mobile apps, and Slack for team Bots. | ChatGPT, Slack and Microsoft Teams. | The Muse app and WhatsApp, with AI glasses announced as coming soon. | On Apple devices, from the side button, the Dynamic Island, Spotlight and context menus, with actions in apps such as WhatsApp; Microsoft Outlook is announced as coming soon. | Word, Excel and PowerPoint built into Copilot now; Autopilot, in private preview, shows up in Teams, Outlook, chats, channels and documents. |
| Distribution and lock-in | Not bundled into a device, an app or an employer. The same governed operator follows you from channel to channel, and is designed to reach more of them; the relationship and the context stay yours. | Reaches people through xAI’s own Grok apps and on X, where you can tag @bot to start work where available. | Built into ChatGPT and rolling out with its Pro and Business Premium plans, so it reaches the people already using ChatGPT. | Available directly in WhatsApp, Meta’s messaging app, and free for most of what people need. | Built into iPhone, iPad, Mac and Apple Vision Pro: the operator arrives with the device. | Built into Microsoft 365, so it reaches people through the work tools their employer already provides. |
| Your data and boundaries | Business, employment and private sources keep separate identity, memory and policy boundaries. Each employer’s confidential material stays isolated, while what Staff learns about how you work carries across. | Your Bots share one cloud computer, with its files and signed-in sessions; xAI advises not to treat separate Bots as a security boundary. | Each dot works from its own cloud computer, building on ChatGPT’s protections, within boundaries you set. | Agent and personal data are housed together in Muse Secure VM; credentials go into secure storage, used without Muse seeing them. | Runs on device or in Private Cloud Compute, where Apple says personal data is not stored nor made accessible to Apple or anyone else. | Lives in your organisation’s tenant, under its data protection policies and IT administration. |
Sources: xAI, “Introducing Grok Bot” (11 Aug 2026); Grok Bot documentation; Grok Bot: Team Bots; Grok Bot: approvals, security and privacy; xAI, “Introducing /goal” (22 Jun 2026); OpenAI, “Introducing dots” (29 Sep 2026); OpenAI Help Center: getting started with your dot; OpenAI Help Center: dots privacy, security and safety; Meta, “Introducing Muse” (8 Sep 2026); Apple, “Siri AI, a profoundly more capable and personal assistant, is here” (14 Sep 2026); Microsoft, “Introducing Microsoft Scout” (2 Jun 2026); Microsoft, “Introducing the new Copilot with Home, Code and Autopilot” (25 Sep 2026).
Each company’s approach as it described it, checked on 9 October 2026; products change.
Product names, logos and brands are the property of their respective owners and are used for identification and comparison only; their use does not imply endorsement. See our terms.
Go deeper
The public page explains the product and the experience. The Capabilities paper describes how state, knowledge, reasoning, authority and execution work together across eleven pages. The Investor Summary sets out the situation, the day and why the market keeps widening, in three. The Architecture Overview is the governed architecture itself, on one page.
Technical
Staff Capabilities →
Eleven pages on state, knowledge, cognition, attention, authority, execution, provenance and runtime.
Investor
Staff in three pages →
A morning that went wrong, a day that did not, and why the market keeps widening.
Architecture
Architecture Overview →
One page · System architecture. Seven layers, one operator — the full governed architecture behind Staff.
The vision
Every era of computing has moved capability closer to the person. The mainframe served an institution, the personal computer served a desk, the network served a household, the phone served a pocket. Each step was real. Each step also arrived with machinery the person was expected to operate: files to organise, applications to choose between, accounts to reconcile, context to carry from one tool to the next.
That machinery has never actually gone away. It was redistributed. The work of being the integration layer — knowing which system holds which fact, restating the same background to a different tool, noticing that something changed and deciding what it means — still lands on the person. Capable models did not remove that work. In many settings they added to it, because now there is also a model to choose, a context window to fill and an output to check.
Staff takes the opposite position. The durable thing is not the model, the device or the interface — all three will be replaced, probably more than once. The durable thing is the operating relationship: what it knows about your situation, how it works with you, and . Those are held outside any component that can be swapped.
This is what makes the coming changes additive rather than disruptive. Models will keep improving. More reasoning will move to private and on-device environments, for reasons of latency and cost as much as privacy. Interfaces will expand from screens to voice, wearables and ambient systems. Under this design each of those arrives as more capability underneath an unchanged relationship, not as a migration you have to survive.
It also sets the limit deliberately. More capability is not more permission. A system that can act on your behalf has to be explicit about what it may know, what it may decide alone and what it must bring back to you — and that boundary should move only when it has been earned, visibly, in a specific domain, rather than granted wholesale because the underlying technology got better.
The direction of travel is consistent even though the parts mature at different rates: from explicit workflows toward quieter orchestration, from connected systems toward ambient context, from scheduled preparation toward continuous readiness, from bounded action toward earned autonomy. That is a direction, not a roadmap with dates.
As technology expands capability, the machinery should increasingly disappear. What remains unique is you.
What you need, when you need it.
