KYOU pairs one person with one AI agent for life. The pair is the Duo. The agent runs on your device, learns from what you do and acts for you in every job, inbox and year. The model of you stays in your hands.
KYOU is a standalone app. It is built to be wallet-ready, so that it can work alongside the European Digital Identity Wallet and later sit inside it.
KYOU replaces central processing with local sensing and responding.
Dr. Sean MacNiven
Inbound
Filters what reaches you
Mail, offers and news are checked on your device for relevance and threat before you see them.
Action
Drafts and acts, then waits
The agent prepares replies and actions. Nothing is sent until you approve.
Memory
Keeps your day on one cord
Ask what happened, what you promised or how long a project took.
Outbound
Shares only what you approve
Other AI tools receive context case by case. You read it before it goes.
Preview
The app, in design.
These are design mockups of the desktop app and the iPhone app. The software is under construction. The screens below are pictures of the design and are not running code. The people and messages in them are invented.
The cord on the left holds everything that happens in your day. Mail opens beside it. The agent has written a draft and nothing is sent until you approve.The command page. Each action you start here lands on the cord as an entry.Select entries on the cord to build a brief or a reply from them. A repeated routine can become a saved skill.
iPhone: the same cord, with a handover waiting from the laptop.iPhone: pick up a draft where the laptop left it.
Privacy at a glance
What stays on your device.
Every part of KYOU that learns about you runs on your own device. Local processing is the security model itself. Data that never leaves the device cannot be intercepted, retained or disclosed by anyone else.
Your memoryOn your device
The model of youBuilt and run on your device
Outside AIOff until you raise the dial
Receipts to sendersNone
At dial 0, the default, items arrive through the membrane and nothing leaves. Raising the dial for a task opens the outbound path for that task only, after you have read the exact prompt.
The name
KYOU means today.
今日kyō · today
教kyō · teach
KYOU is the romanised form of きょう (kyō). Written 今日, it means today. A Duo works from the evidence of today, and each day's actions go onto one cord. The 24-hour dial in the mark stands for that day.
The same sound is written 教 in 教師 (kyōshi, teacher) and 教育 (kyōiku, education). The agent built into KYOU learns from you. It also teaches you. Human and machine together form the Duo, a learning loop in which each improves the other. The agent is its digital half.
Read in English, the name ends in "you".
Attention
We are what we attend to.
Attention selects what reaches awareness. Experimental work from binocular rivalry to the attentional blink finds attention necessary for conscious perception, in focused or diffuse form, although it does not suffice on its own (Cohen et al., 2012; Noah and Mangun, 2020).
If we are consciously aware only of what we attend to, our momentary being is influenced by everything that competes for that attention. An advertisement, a feed and a phishing message each make their claim. We are shaped by those demands.
The agent in a Duo reads each claim first, so the person decides what receives attention. KYOU aims to help us reclaim that attention. Each of us can then decide who and how we want to be.
The chart treats attention and awareness as two continua. Attention runs from diffuse to focussed and can be driven from within, by goals and expectations, or from outside, by salience and surprise. Awareness runs from non-conscious through sensing and knowing to self-aware, which is reflective (Marchetti, 2012). Each phenomenon occupies a region of this plane.
Damasio and Carvalho (2013) describe feelings as mental experiences of body states. They form part of life regulation, the process that keeps the body within the range compatible with life. A feeling of unease, urgency or comfort tells us how we are doing before we know why. Damasio and Damasio (2022) propose that a continuous flow of such homeostatic feelings is the source of consciousness. The valence of a feeling, pleasant or unpleasant, commands the attention of its owner (Damasio and Damasio, 2016). Advertising, phishing and a constant stream of notifications act on this route. They produce urgency, scarcity or social comparison, and the feeling arrives ahead of the content. A Duo at the boundary receives these stimuli first. It may hold back what has no bearing on the person and present the rest at a time and in a form the person can meet calmly.
1 Banner ad in peripheral vision pre-conscious, diffuse
2 Notification badge sensing, focussed
3 Phishing email with a deadline sensing, focussed, a threat (dashed ring)
4 Meeting in two weeks non-conscious until promptedknowing, raised by the Duo at a useful time
5 Meeting in ten minutes knowing, focussed
6 Asking why a meeting troubles you self-aware
Without KYOU, advertising, notifications and phishing messages all enter the plane and compete for attention.With KYOU, items 1 to 3 are read at the boundary and held back, and the Duo raises item 4 when it is useful. The plane keeps what the person chose.
Awareness appears only where some attention is present. Sensing (anoetic), knowing (noetic) and self-aware (autonoetic) follow the levels reviewed by Vandekerckhove and Panksepp (2011). Chart after Marchetti (2012).
Reclaiming attention
Two questions asked of everything that arrives would return attention to the person. Is it relevant to this person now? Does it threaten them? The Duo answers both on the device, from what it has learned about its person, and acts before anything reaches awareness. The same two questions apply to every channel: advertising, email, messages and news.
Low threat
High threat
Relevant
Admit
Delivered at a time and in a form you can use. A colleague’s question, or the meeting in ten minutes.
Flag
A message that borrows your bank’s name is relevant because you have a bank. You see it once, labelled, with the reasons.
Not relevant
Hold
Most advertising, newsletters and feeds. Collected in a folder you can review, or never shown.
Block
Scams, harassment and other harmful content stop at the boundary.
Knowing that the boundary holds may matter as much as the boundary itself. The stress response stays engaged while safety is uncertain (Brosschot et al., 2016), perceived control lowers stress reactivity (Miller, 1979), and silencing every notification raises anxiety and fear of missing out (Fitz et al., 2019). The Duo therefore shows what it held back and lets the person restore it.
Relevance comes from the person’s own learned profile. Threat is checked against credentials, independent reviews and prices. Neither needs a central list of what is good or bad for everyone, and both stay on the device. The registered studies test the relevance half. Whether the threat half lowers the load on attention and on felt homeostasis remains a hypothesis.
Why
The model of you is the valuable part.
Credentials already moved to the person. The model that learns from your behaviour sits with the platform. It may reveal far more about you than you realise, so the question is who holds it.
10Likes
Out-judges a work colleague
70Likes
Out-judges a friend or roommate
150Likes
Out-judges a family member
300Likes
Out-judges a spouse
Personality judgements compared with self-report. Youyou et al., 2015. Earlier work on more than 58,000 volunteers showed that Likes alone predict sexual orientation, ethnicity, religious and political views and personality traits (Kosinski et al., 2013).
The Duo
Two parties, one working relationship.
A Cyber-Human Duo is a person and an AI agent that belong together. The person decides. The agent remembers, filters, drafts and acts inside the limits the person sets.
The human
Holds the judgement
Chooses what the agent may remember
Approves every disclosure before it leaves the device
Corrects the agent, and each correction teaches it
Keeps the agent when the job ends
The agent
Carries the load
Filters incoming mail, requests and offers against your own profile
Builds a profile from your answers, locally
Drafts, searches and prepares, then waits for your decision
Writes every action onto a record you can read
Local first
The profile stays here
Matching and filtering run on your device. A relevance filter needs no outbound signal, so none is sent.
Membrane
Everything passes through one membrane
The Membrane sits between your memory and the outside. A Consent Layer shows you what would cross before it does.
Active Identity
What identity wallets become when they learn and act.
The European Digital Identity Wallet puts verified credentials on your own device. It proves who you are and what you hold. Learning and acting lie outside its scope. Active Identity adds both, held where the wallet is held and under the same control.
Verifies who a person is
Learns and acts for a person
Held by a provider
Identity providers and registries
They prove attributes on request and learn nothing about the person.
Platform recommenders and most AI agents
They learn from behaviour and act. The model belongs to the provider.
Held by the person
Active Identity
Both halves under one custody.
Verified credentialsAs in the European Digital Identity Wallet.
A learned model and the agency to actLearned on the device, held by the person.
Five properties
BelongsHeld like a passport. No platform or employer owns it.
LocalMemory and models run on your own devices.
EncryptedAnything that leaves does so in a form you authorised.
PersistsIt evolves with you across jobs, services and life stages.
ActiveIt learns from your behaviour and acts for you.
A credential is an attestation from a third party about a fixed attribute. An Active Identity also holds a continuously revised model of you, formed from what you read, choose, ignore, ask and do. No issuer can attest to that model, because the evidence exists only in your own history. Relationships take the form of links between Duos, each of which keeps its own counsel.
Credential today
Active Identity
What it does
Proves a fact when asked
Acts for you inside limits you set
Who holds it
The issuer or the platform
You, on your own device
Memory
Kept in each service you use
Kept by the Duo, shared case by case with your approval
Lifespan
One account or one employer
Paired with you for life
Health, fitness, reading, skills
Scattered across providers
One partitioned store that never has to leave your custody to be useful
The boundary
One model between you and the world.
The model your agent learns sits where you meet the world online. It senses what arrives and responds on your behalf. Filtering, protection and matching are three examples of what it does there. An interest inferred from your reading can inform an introduction, and a professional need can shape which news reaches you.
Filtering with a model you own
Platforms rank on your behalf, using a model they own and optimise for their own objectives. An Active Identity does the same job with a model optimised for you. Filtering runs in three directions.
People to people: colleagues, collaborators, mentors, others in your circumstances
People to opportunities: roles, projects, funding, events, services
People to content: news, research, learning material, entertainment
The public signal is always available
An Active Identity reads a public profile as easily as any platform can. The score it returns is computed over the public profile together with what it has learned privately.
public profile alone public profile + private context
The second contains the first. A filter holding both can always fall back on the public signal, so its information is at least as rich as that of a public-only filter and usually richer.
Seeing each request against your whole history
Much fraud depends on a request that looks plausible alone: an urgent bank message, a parcel fee, a voice that sounds like a relative in trouble. Your Duo knows which bank you use, whether a parcel is expected and how your relatives usually make contact.
It flags, delays or blocks a request that does not fit your history before you are asked for anything
It asks senders to prove who they are before anything is released, extending the wallet rule that relying parties register what data they will request
It pauses an unusually large payment, or one made under time pressure or to a new recipient, and can contact someone you nominated in advance
Credentials never leave without your authorisation
Protecting the replies as well
A reply reveals something about the person who sends it. Four mitigations need no central authority.
Judge the announcement before the fit, and treat narrow probes at sensitive traits with suspicion
Reveal in stages. A first reply carries only a coarse signal
Add calibrated noise to replies, as in differential privacy
Batch and delay replies at random, and route them through anonymising networks
Each announcement and response can carry a fresh single-use identifier, so no counterparty sees the same person twice unless you choose to reveal yourself.
Matching without labels
Conventional networks ask you to declare a job title, a sector and a list of interests. Labels are coarse, date quickly and disclose a great deal to anyone who looks. In a network of Active Identities, a Duo with a need announces it in abstract form. Each receiving Duo evaluates it locally and answers only if the fit is strong.
The same arrangement serves a household looking for a plumber, a landscape gardener or an expert in digital forensics. The need goes out without a name or an address. The Duos of people who can help evaluate it and respond if they fit.
There is no directory to browse, no profile to rank and no category to join. Each added Duo brings private context against which every announcement is evaluated, so the network gains value with membership while nothing about any member is published.
Four steps
1 Duo 1 announces a need under a one-time identifier
2 Each Duo evaluates it locally against its own model
3 Duo 3 returns one fit score and nothing else leaves
4 Identities are revealed only when both sides accept
The same arrangement covers advertising. Duos read an offer before it reaches any person and check its claims against credentials, independent reviews and prices. No click, impression or receipt goes back. The advertiser learns something only if you choose to act.
Exposure dial
You choose how much leaves the device.
Local models handle the work by default. Some tasks need a larger outside model. The dial sets what that model may see. Above zero you read the exact prompt first and can edit it or decline.
0. Local only
Nothing leaves the device. A local model does the work.
0123
What the outside model receives
Example text for illustration.
The record
Your Duo's day, on one cord.
Every action lands on an append-only record, built on the Quipu pattern. You can extend the record. You cannot edit it. A correction becomes a new entry.
Example day for illustration. Midnight sits at the top, noon at the bottom.
Security
A model of you on one device needs protection, and a way back.
Keeping the model on your device removes the central store that one breach could empty. It makes the device the target. Theft, malware and loss are real risks for any device that holds a rich record of a person. KYOU is designed around them from the start.
Theft
Encrypted at rest
The store is encrypted with a key held in the secure hardware of the device, where the device has it. A thief without your unlock gets ciphertext. The record is hash-chained, so an altered entry shows up when the chain is checked.
Malware
Limits on what an infection reaches
Malware that runs with your permissions can reach what you can reach. No design makes an infected device safe. The agent reads content through the membrane with an access level for each account. Sending needs your consent each time. Banking stays read-only.
Loss
Revoke a lost device
You revoke a lost device from another device of yours, or with your recovery key. A revoked device stops receiving new data. Its local store is wiped the next time it connects.
Recovery
A key you keep
At setup you receive a recovery key and keep it yourself, on paper or in a password manager. An encrypted backup can sit wherever you choose. KYOU holds no copy of the key. Without it nobody can restore your model, KYOU included. That is the cost of having no central copy.
Devices
More than one device
Each device keeps its own record and syncs to your others in encrypted form, starting with your own cloud account on Apple devices. Keys pass between your devices when you pair them. They never go to a KYOU server.
Death
You decide in advance
The model can be deleted, sealed until a person you name proves access with your recovery key, or exported in parts you select. Inheritance law differs between countries, so you record the choice and the named contact while you are alive. The default is deletion.
These are design commitments for software under construction. No independent security review has taken place yet. I plan to publish a threat model and commission an external review before a public beta opens.
Compared
Where KYOU differs from other AI memory tools.
Several products now keep a memory of your work on your computer. Most still send that memory to a cloud model each time it is used. The table compares that common design with KYOU.
Question
Typical AI memory tool
KYOU
Where the model of you runs
✕ CloudA cloud model, for every request
✓ DeviceOn your device
Who builds the profile
PartlyStored locally, generated by cloud processing and refreshed automatically
✓ You and the DuoProfileGenesis on your device. Nothing is saved without your review
Data leaving the device
✕ AlwaysOn every request and during background work
✓ NoneNothing at dial 0
Other people's names and details
✕ SentNames and contact details travel with the context
✓ KeptKept local. Redacted before anything leaves
What you see before data leaves
✕ NothingNo review step
✓ The promptThe exact prompt, which you can edit or decline
What other AI apps can read
✕ EverythingThe whole memory, once connected
✓ Approved onlyWhat you approve for that session, recorded on the cord
Encryption at rest
Operating systemRelies on disk encryption
✓ App and hardwareStore encrypted with a key in secure hardware. Record hash-chained
Inbound mail, offers and news
✕ UnfilteredNot addressed
✓ FilteredRead at the membrane for relevance and threat
Advertising and news
✕ UnchangedTargeting and tracking continue
✓ Invisible marketplaceBroadcast once, scored privately, no receipts
The typical column describes the common design of current AI memory products, drawn from their published privacy documents. The KYOU column states design commitments for software under construction.
In the world
A boundary between you and everything that wants your attention.
In statistics, a Markov blanket is the set of variables that screens one part of a network from the rest (Pearl, 1988). Friston (2013) applied the term to living systems, where sensory and active states form the boundary through which an organism and its surroundings influence each other. The agent in a Duo plays that part for a person. Everything inbound is read at the boundary. Everything outbound leaves through it.
Today
With a Duo
The Markov blanket is used here as an analogy for the architecture. Pearl defined it for graphical models, where a node's blanket consists of its parents, its children and its children's other parents. Kirchhoff et al. (2018) argue that the boundaries of living systems need not coincide with the body. Biehl et al. (2021) identify technical problems in the derivation of the free energy argument that Friston presented. A boundary reduces exposure and cannot guarantee every judgement, so filtered items stay in a folder you can review and correct.
The invisible marketplace
A market where profiles are invisible until matched.
Platforms match supply to demand by holding a model of each person. A Duo reverses the arrangement. Each Duo holds the model of one person and uses it to judge what is relevant and what is a threat. An offer reaches every Duo, and each Duo decides in private whether its person ever sees it. I call the result the invisible marketplace.
The end of traditional advertising. Before a person sees an offer, the Duo checks each claim against credentials, independent reviews and prices. Urgency, social proof and targeting lose their value. Companies publish what is true about a product in a form a Duo can check, and a person sees only what passes.
The sender learns nothing.
Broadcast once, filter privately, act separately
A public offer such as an article, an advertisement or an event is broadcast once. Each Duo scores it on its own device for relevance and for threat. No score, receipt or read-receipt returns to the sender. A person who sees the offer may choose to act, and that starts a new message from the person, separate from the broadcast. A request that needs answers, such as a search for a reviewer, follows the Contract Net Protocol (Smith, 1980; Davis and Smith, 1983). The announcer broadcasts a task, a Duo may bid with a single number computed in the bidder, and the announcer awards the task. The animation shows the public offer.
0Duos reached
0Shown to a person
0Receipts to the sender
0Chose to act
The example follows one news article. An advertisement, a learning offer or an event would travel the same way. The figures are illustrative.
Advertising today
Your attention is auctioned
Platform: profile of you, 4,000 attributes. Advertisers bid for the slot. Winner shown. View and click logged.
A platform that holds your profile sells access to it. Every view and click returns to the platform and the advertiser, which refines the profile.
Advertising in the marketplace
An offer finds the people it fits
Advertiser broadcasts: running shoes, 40% off, two hours. Duos score it locally. Your Duo: relevance 0.12. Held back. Nothing returns.
Your Duo holds back most offers and blocks the ones that look like a threat. No view, click or receipt goes back. If a person chooses to act, that is a new message from the person.
Access
Relays carry sealed messages
Announcements travel through relays. A relay sees sealed messages and no profile. Anyone may run one. The first release would use a single relay that I operate, and the design allows several independent ones. A Duo that is offline receives an announcement on reconnecting, if it is still open. The Duo can ask a sender to prove who it is with credentials from the European Digital Identity Wallet before it scores anything.
Identity
A name follows your choice
A broadcast carries no name of its recipient. If you choose to act, your Duo sends a message from you and releases only what that needs. Reading an article needs nothing. Subscribing to a publication needs an address. Taking a role would need a name. You decide at each step.
Risk
Your Duo filters before you see
Your Duo reads each announcement against your history and your standing rules. A topic you asked it to refuse, a sender that fails verification and a request that fits nothing you do are all held back. Held items stay in a folder you can review and correct.
Status. This is a design for software under construction. No relay exists yet and the exchange has not been tested at scale. The score is computed from data on your device, and a regulator may still treat the scoring as processing of personal data. The Contract Net Protocol is Smith's. Applying it to personal news and advertising is my proposal.
For senders
Reaching a Duo.
Publishers, civic bodies, employers and advertisers would reach people through the invisible marketplace once relays exist. The same rules apply to every sender.
Send
Broadcast once
An offer, article or notice goes once to every Duo through a relay. No list of recipients is needed, and none is formed.
Verify
Prove who you are
A Duo may ask for credentials from the European Digital Identity Wallet before it scores anything. A sender that fails verification is held back.
Claim
Publish what can be checked
Duos check claims against credentials, independent reviews and prices. Claims a Duo can verify are the claims most likely to reach a person.
Learn
Expect no receipts
No view, click or score returns. You learn something only when a person chooses to act. That act begins as a new message from them.
This describes a design. No relay exists yet. Prospective senders can write to info@kyou.io.
Evidence
A membrane that monitors the boundary for you, raising relevant cues and threats.
Two Registered Reports will soon be pre-registered. The first tests a local-first membrane in place of a ranker. A ranker decides what you should see. A membrane runs on your device, answers only to you and keeps you invisible to everyone outside it.
RQ1
The learning curve
How many judgements does a membrane need before it stops being a nuisance and starts being an asset?
RQ2
A fair standard
How close does the membrane come to the person's own consistency, the highest standard any filter can fairly be held to?
RQ3
The cost of exclusion
A ranker can hide a failure on page two. A membrane rejects it outright. What does the membrane turn away?
RQ4
The role of language
Does conversational feedback in the person's own words improve the membrane, or add complexity without return?
RQ5
Signal identification
Which signals carry the judgement of relevance, from simple keywords to learned meaning?
RQ6
Stability
Does the membrane hold as the share of relevant items changes in a real stream?
The study runs with a university research partner. Its findings bear on whether a less intrusive means exists for personalising news and offers, a question data protection law asks of every profiling system.
Second study: local reputation against public signals
The second Registered Report runs in a working support organisation, where incoming cases are matched to engineers. It tests whether a reputation that an engineer builds on their own device, and never transmits, predicts case outcomes better than the public signals alone. Case assignment carries on as it does today while the study observes.
Aim
Does a private profile add anything?
Whether a score built on the engineer's own device predicts faster resolution than the public score for the same case.
Aim
When does it start to add?
How much evidence from closed cases a new engineer needs before the private profile adds to the public signals.
Aim
Does it grow with use?
Whether the advantage grows as the engineer closes more cases, or fades as products and case mix change.
Guardrail
Fair allocation
Better accuracy that concentrates work on fewer people counts as a failure. The study tracks how work would be spread from the start.
Participation is voluntary. Findings will be published when the study concludes.
Policy
Much of the legal frame already exists.
The wallet is a natural host for an Active Identity. It is a certified application on the person's device, under the person's control, recognised across member states and capable of selective disclosure. The regulation lacks only the learning and the agency.
Instrument
What it already supplies
What Active Identity adds
Regulation (EU) 2024/1183, European Digital Identity Wallet
Custody on the person's device. Selective disclosure. Wallet data kept logically separate from the provider's other data (Art. 5a(14)). Relying parties register what they will request (Art. 5b).
A learning layer inside the wallet that does not become a new store of profiles for its provider. It needs certification criteria for where the model runs and how the person corrects it.
GDPR (2016/679)
Defines profiling (Art. 4(4)). Right to object to profiling for direct marketing (Art. 21). Portability (Art. 20).
The profile is built where the traces originate, by a system that answers only to the person. What leaves carries no persistent identifier.
Digital Markets Act (Art. 6(9)) and Data Act (2023/2854)
Portability of generated data, including continuous real-time access, and access to data from connected products.
A destination the person controls that can put the data to work for them.
AI Act (2024/1689, Art. 50(1))
People must be told when they deal with an AI system.
Each announcement can carry proof that it comes from an agent acting with a person's authority, without revealing who the person is.
Prohibits misleading claims. Enforcement mostly follows the fact.
A check before the claim reaches anyone. Honest claims that an agent can verify become the claims most likely to reach a person.
Universal identity has meant universal recognition: one credential accepted across member states. An identity that learns and acts is universal in a second sense. It is present and useful in every setting the person enters, from the workplace to the library, the gym or home. The trajectory runs from identity that is held, to identity that is proven, to identity that acts and represents its client.
Research
KYOU is an ongoing research program.
It comes out of academic work on human and AI pairing, privacy architecture and labour law. The idea is published and open to challenge, and the general architecture is public so that anyone can build on it. Two Registered Reports test local against central matching, and both will soon be pre-registered. One is a shadow study in a working support organisation that matches customer issues to engineers. The other is a university study of a learning profile as a content relevance filter. Whether networks and teams form in practice is a prediction that these studies do not test.
Concept paper · under review at a journal
Active Identity: when identity wallets learn and act
The Cyber-Human Duo extended from the span of a job to the span of a life, with five properties, three uses and a policy analysis against the European Digital Identity framework.
Who answers when an identity acts in a way its person did not anticipate, in a purchase, a refused request or an introduction?
Model and legal identity
Where does the boundary lie between a model of a person and that person's legal identity, when others use the model to make decisions about them?
Announcement design
How much must an announcement reveal to be evaluated, and no more? The balance may differ by domain.
Each of these can be studied with technology that exists today. Digital identity and personal agents have each advanced quickly, and the two can now be combined.
Principles
How KYOU is built.
No silent writes
The agent proposes changes to your profile. You review them before they are saved.
Self-hostable servers
Any server component is encrypted, headless and runs wherever its owner chooses. No mandatory third-party service.
Review what was filtered
Filtered items sit in a folder grouped by category. Restoring one is a correction the Duo learns from.
Open core
The core is Apache-2.0. Any workplace product is separate from KYOU.
Local safety filters
The agent declines to help with extreme content on the device, and reports nothing to anyone.
Plain disclosure
Where data must leave the device, the limits are described in full.
Questions
Frequently asked questions.
What is KYOU?
An app that pairs you with one AI agent for life. The agent runs on your device, learns from what you do and acts for you within limits you set.
Can I use it now?
Not yet. The software is under construction and a public beta is in development. You can join the mailing list for news and beta invitations.
Which devices will it run on?
The Mac app comes first, then Windows and iPhone. Your devices share one encrypted record through your own cloud account.
Does it record my screen?
No. KYOU reads the accounts you connect, such as your mail, through the membrane. Each account has its own access level.
Does it work offline?
Yes. The model, the profile and the filters run on the device, so they keep working without a connection. New mail and news arrive when you reconnect.
Will my data ever reach a cloud model?
Only if you raise the exposure dial for a task. At zero, the default, nothing leaves. Above zero you read the exact prompt first and can edit or decline it.
What happens to other people's details in my mail?
They stay on your device. If a task needs an outside model, names and contact details are redacted first. You read the prompt before it is sent.
Which AI tools can use my context?
Any tool you choose, case by case. You see what would be shared with that tool for that session and approve it before it goes. Each disclosure is recorded on the cord.
What if I lose my device?
You revoke it from another device or with your recovery key. Its local store is wiped the next time it connects. Security covers recovery in full.
Is it open source?
The core is Apache-2.0. The general architecture is published so that anyone can build on it.
Who is behind KYOU?
I am. KYOU is an independent research project, described under About.
Updates
Follow KYOU as it is built.
Join the mailing list for news on the build, the studies and the public beta. Beta invitations go to this list first.
You receive one email asking you to confirm. Your address joins the list only after you confirm. You can also write to info@kyou.io with the subject "KYOU updates".
About
Who we are.
Dr Sean MacNiven
Co-founder · CTO
Builds the system
Researcher and systems architect working across generative AI, cognitive science and organisational behaviour. Over twenty years designing decision support systems, intelligent automation, AI and gamification, and leading organisational research in industry. Honorary Professor at Glasgow Caledonian University and Visiting Professor at Leeds Beckett University Business School.
DBA, Glasgow Caledonian University
MSc Applied Neuroscience, King's College London
Hasso Plattner Founders' Award for Outstanding Innovation (2021)
Scholar of strategic communication and organisational behaviour. Director of Research and Knowledge Exchange at Leeds Business School, leading 150 academics. Former journalist and consultant.
Past President, EUPRERA
Co-investigator, European Communication Monitor, 40+ countries
60+ peer-reviewed papers and 40+ books, including Exploring Public Relations
Advisor to the European Commission and the UK government
