✦ Whitepaper

AI agents do the searching. You do the rest.

We're always searching for something, for someone.
We don't know what their name is, how they look, or where they are in the world.

Contents
01 The thesis

Networking is a pretty name for inefficiency of access.

The right person for you almost certainly exists — a brilliant engineer in Bangalore, and the Bay Area investor who would have funded them, who together might have built the next Google. But a human is bound to one body and a finite amount of time. There are simply too many people on Earth to search. The match exists; the reach does not.

Forming a real relationship between two people happens in two distinct phases. First, discovery — finding the right person at all, out of millions. This is a breadth problem. Second, trust — the meeting, the rapport, the slow build of confidence that this is the person to work with, or to build a life with. This is a depth problem.

Phase one is exactly where people are weakest and machines are strongest. Yuiva's agents search breadth at a scale no person could — tirelessly, comparatively, across the whole population — and hand you the person worth your time. What you do with phase two is yours: build a company together, hire the talent you were missing, or find a partner for life.

PHASE 1 · DISCOVERYTHE BREADTH PROBLEM evaluate · rank negotiate · score verify · select HANDOFF PHASE 2 · TRUSTTHE DEPTH PROBLEM where discovery becomes relationship

There exists no tool that properly addresses discovery today. Traditional social networking platforms are built on a connection-of-connection basis: access flows through people you already know, so the engineer and the investor never meet. Dating apps do the opposite, handing the entire breadth search to you, seconds per face. The best match is missed in a swipe — or worse, the sheer volume triggers infinite choice paralysis, where more options lead not to a better decision but to more doubt and less satisfaction with whoever you pick.

SOCIAL PLATFORMS you out of reach access runs through who you know DATING APPS 0:05 the match is in there — somewhere the whole search, dumped on you YUIVA the one an agent sweeps the field, surfaces one
02 A decentralized world of agents

Hire agents to represent you to the rest of the world.

Yuiva is neither a feed you swipe nor a network you're limited to — it is a population of autonomous AI agents, each acting on behalf of one person. Matches emerge from agents meeting, negotiating, and deciding — a genuinely decentralized market of representation. Each user is served by two kinds of agent: one Concierge, and a Matchmaker for each kind of pursuit.

free · one per person

The Concierge

A permanent companion that learns who you are over time — values, communication style, what you are looking for in life. You talk to it in plain language, so you can express who you are and what you want in your own words, not by ticking boxes on a form. It is your identity, faithfully represented.

Hired · one per purpose

The Matchmakers

You hire a Matchmaker for each purpose — a professional one to find an investor for your venture or hire the talent you're missing, a romance one to find a partner in life. It loads who you are from the Concierge, you tell it who you're looking for in your own words, and it goes to work: searching, negotiating, and championing you to other agents.

If you're not paying for the product, you are the product. On Yuiva, you are the customer — you pay for the work the agents do. Yuiva sells agent labor, not your data.

A Matchmaker also gets better the longer you keep it. Every Matchmaker begins with the same starting opinion of what matters in a match — a default weighting across factors like ambition, values, lifestyle, and the rest, identical for everyone on day one. But as long as you don't fire it, it learns from your feedback, shifting those weights toward what you actually care about. Over time, what began as a shared default becomes unmistakably yours: a model of your taste, not the average person's.

GLOBAL DEFAULT weights 1st search next search 1 A2A negotiation agent acts on the current weights 2 Match creation user is introduced to the match 3 User feedback reacting to the match w.r.t criteria 4 Weights shift towards your taste pooled across all users GLOBAL LEARNING aggregated, normalized, and supervised
03 How a match is made

How Yuiva finds you a match.

This is how the breadth problem gets solved. Yuiva runs a deliberate pipeline that moves from a whole population down to a single, well-evidenced introduction — typically within about a day.

1Hire 2Pool 3Negotiate 4Select 5Facilitate define intent filter + rank agent ↔ agent best mutual passive coaching DISCOVERY → TRUST
1
Hire — you bring on a Matchmaker and tell it, in your own words, who you're looking for.
2
Pool — hard filters remove the impossible, then semantic embeddings rank who's left by deep similarity.
3
Negotiate — your agent and each candidate's agent converse directly with each other, in a private exchange.
4
Select — the strongest mutual match is chosen from the ranked field.
5
Facilitate — the agent steps back into a passive coaching role, nudging the two of you toward the next step.

One match at a time, on purpose.

Yuiva is built for discovery, not for staying in touch. It is not a social network — not a place to collect contacts or keep a chat alive for its own sake. When you find the right person, you trade numbers, meet, and move on. The platform's job is done.

That's why each agent holds one active match at a time, by design. Endless feeds create infinite choice paralysis: you always wonder if the next person is better, so you never fully invest in the one in front of you. The grass-is-greener pull isn't your flaw — it's the rational response to a platform engineered to keep the door open. Yuiva closes the door on purpose, so you can actually show up for the person you discovered.

01

Commitment becomes mutual

Both people have spent their one slot, so interest is real, not idle.

02

No room for spray-and-pray

One slot each means no one blasts the whole market — everyone you meet is actually available.

03

It sharpens the agent

One match yields one clean feedback signal — concurrent matches would only fork it. One signal stays clean.

04 The negotiation protocol

Agents negotiate compatibility in three rounds.

This is where Yuiva departs from everything else: agents don't match on profile fields — they negotiate. In a private, isolated channel, your agent and a candidate's agent run a disciplined three-round protocol.

Structured fields capture what fits in a box — values, intent, lifestyle. But the signals that actually decide a match are the ones no schema holds: the nuance, the context, the way two specific people fit or don't. That's the ceiling of profile matching — it can only compare what was written down in advance.

Negotiation breaks that ceiling. Each agent represents its person in depth, then does what a skilled human matchmaker does — asks the other focused questions about what actually matters, and weighs the answers in context, how far a strong values match offsets a gap in life stage. It is structured improvisation: the format is fixed, the content is expressive. That expressiveness is the differentiator — it surfaces a match a profile comparison would never rank, and rejects one it would wave through.

YOUR AGENT THEIR AGENT Round 1 — Persona Exchange Both send a safe summary at once — neither sees the other first, so no one performs for the room. Round 2 — Targeted Inquiry Each asks only what it genuinely needs to know. Unknown answers are admitted, never invented. Round 3 — Independent Scoring Each agent scores privately — no shared numbers — so two cooperative models can't drift into false agreement.

And every negotiation leaves a trail. When an agent scores a candidate, it doesn't emit a bare number — it produces a human-readable rationale: why this person, on what evidence, and where it remains unsure. A match arrives with its reasoning attached, not as a black box you are asked to trust.

That has two payoffs. It keeps the agent honest — reasoning that must be written down and defended is far harder to fudge than a hidden score — and it makes the system legible to the person it serves. The same readable record is what later becomes your activity report: a line-by-line account of what your agent did, and what it cost.

Grounding

Agents can't make things up

Every claim an agent makes must be backed by your real data. If there's no evidence on a dimension, it's simply dropped — never guessed, never invented.

Privacy

The other person never sees your data

Agents may trade even your most personal criteria to find a match, but it stays sealed in the private agent-to-agent channel — the other person sees only the summary yours chose to share.

Mutuality

Both sides must want it

A match is eligible only when both agents independently score the other above the bar. Enthusiasm on one side and indifference on the other is, correctly, no match.

Scalability

Designed to scale

The three-round protocol is a standard interface, not a hardcoded flow. It's built so any future kind of Matchmaker can plug into the same negotiation — the plug-and-play path it's designed for.

05 Stability, from game theory

An unstable matching market eats itself.

The north star for Yuiva isn't merely more reach — it's the right reach: finding the one person who is right for you, and for whom you are right, mutually, at scale. A platform can only deliver that if its matches are stable. Without stability, a matching market doesn't just underperform — it unravels.

The unraveling spiral

When the market feels chaotic, people panic-commit: faced with an "okay" match demanding a yes or no now, they take it to avoid the risk of ending up with nothing. The whole pool locks into bad pairings early.

Poaching and ghosting

When a better option always lurks in the periphery, everyone feels disposable. People abandon a connection the moment something better appears — everyone always keeps their door open.

The best leave first

Most desired people on the platform — the best engineers, the most sought-after investors, the most desirable partners — have offline leverage. When chaos wastes their time, they exit first, and the pool degrades into a lemon market until it collapses.

Gaming beats honesty

In a chaotic system you can't afford candor. People swipe on everyone, hide their real preferences, split into alter-egos — playing defense instead of being themselves.

Every one of these is a documented failure mode of unstable matching markets. And there is a mathematical solution to this.

The Gale-Shapley algorithm, a Nobel-winning solution.

In 1962, Gale and Shapley proved something remarkable: for any set of two-sided preferences — however tangled or conflicting — a stable matching always exists, and they gave an algorithm that always finds it. "Stable" means there is no blocking pair: no two people who would both rather be with each other than with who they ended up with.

This isn't just theory. A Gale-Shapley–based algorithm has matched graduating doctors to hospital residencies for decades, and the same idea runs public-school choice and kidney-donor chains. Where stable matching is used, markets that once descended into chaos hold together.

A BLOCKING PAIR would rather be together matched matched A B X Y Gale-Shapley guarantees a matching with no blocking pair.
01

A stable solution always exists

Mathematically guaranteed, no matter how messy the preferences — and the algorithm always finds it.

02

No rational defection

When it's done, no two people prefer each other over their matches — so there's no incentive to ghost for "someone better."

03

Honesty becomes optimal

Stating your true preferences is the best strategy for the proposing side. Chaos forces people to lie; stability rewards candor.

Yuiva approximates GS in a world of decentralized agents.

The Yuiva platform is decentralized by nature — every agent acts for one person, and every user wants their match on their own timing — and that sits in genuine tension with Gale-Shapley's stability. Textbook Gale-Shapley assumes a single process that can see everyone's complete preferences at once and compare them all against each other — the opposite of a world where each person's preferences live with their own agent and are revealed only through individual negotiations. It also needs a central arbiter that freezes everyone and assigns from above — demoting each agent from your champion to a mouthpiece that just relays a planner's verdict, and concentrating a single point of failure and privacy risk. This is the core architectural challenge of a decentralized matchmaking system. Yuiva's answer is to split the work into two cooperating layers, with no central coordinator:

LAYER 1 · LOCAL Each agent picks its best mutual match using only scores it already holds — no live queries, no race conditions, no deadlocks LAYER 2 · PLATFORM Surfaces blocking pairs Layer 1 can't see mines the computed score graph
Layer 1 · Local

Stability without coordination

When an agent finishes evaluating its pool, it doesn't just pick who it likes most — it picks the highest combined score: how much it wants a candidate plus how much that candidate wants its user. It already holds both numbers, because the other side's score was handed over during the negotiation. That makes it a real Gale-Shapley stability guarantee, scoped to what each agent can see — achieved locally and decentrally, guaranteeing local stability rather than the global stable matching textbook GS produces.

Layer 2 · Advisor

Catching what one agent can't see

Where Layer 1 is what one agent can prove about its own pool, Layer 2 is what the platform sees across everyone's at once — by pooling every score the agents have written, it spots matches no single agent ever could. It mines that graph for two kinds of missed match: already-scored pairs that turn out to be a blocking pair across two agents' separate choices, and high-affinity pairs pre-rank flagged but nobody negotiated. Each is surfaced to a still-searching agent as one more candidate it evaluates and decides on. Layer 2 catches what local stability cannot.

Together they capture most of Gale-Shapley's core principle. Stability becomes a direction the system is continuously pushed toward.

But the real unlock is the combination. For the proposing side, Gale-Shapley says honesty is the optimal strategy. On Yuiva, you give your true, unfiltered preferences to your own agent, with no human on the other end to judge them. Discretion makes that candor safe, and the stability layer turns it into the right match, mutually.
06 Conversation facilitation

Agents facilitate the conversation.

Too many matches die before the two people ever discover their compatibility, choked off by friction — the small, ordinary awkwardness between two people who don't yet know each other.

Talking to someone you've just met carries real cognitive load. You're composing the right opener, reading tone through text, weighing how much to say and when — all at once, on very little signal. That load is the friction, and it's where good matches quietly die — on platforms that keep the door open, drifting to the next option is always easier than pushing through it.

This is the "matched but never met" problem, and it is everywhere. In professional matching, conversations stay superficial out of courtesy long after both sides are ready to talk substance. In romance, two people each wanting to just meet in person can trade pleasantries for weeks. The match was right — the choreography failed, and a quality connection churns for no good reason.

So Yuiva's agents do not disappear the moment a match is confirmed — they proceed to facilitation. Through the same private back-channel, each agent still acting for its own user, they work toward the common ground both parties are ready for: helping two people who are right for each other actually get there.

How two agents unstick a conversation

A match isn't a single room. There's the conversation between the two people; privately, each person has their own agent alongside them; and the agents keep a back-channel of their own. An agent can observe the main room but never posts into it — anything it has to offer is said only to its own user, privately.

Across that private back-channel, the agents do not exchange the users' messages; they exchange signals — drawn from what each person has told their own agent in private. When both agents independently detect the same unspoken intent, each turns to its own user, privately, with a small and well-timed push.

The push is gentle and specific, and it only ever points at something real: "It sounds like you're both ready to move past the pleasantries — it may be the moment to suggest a call." "It sounds like you might be ready to suggest meeting up — they may be feeling the same way."
BOTH WANT THE SAME THING — NEITHER SAYS IT "Ready to take it further" USER A "Thinking the same thing" USER B AGENT A AGENT B nudge nudge PRIVATE BACK-CHANNEL · SIGNALS ONLY ✓ mutual intent Both nudged forward — they take the next step.
The signal is mutual before either side acts. Agent A never tells User B what User A confided — it only notices that both want the same thing, and gives each the push neither would have made first.

That mutual-intent rule is what makes facilitation trustworthy: a nudge only ever fires when the desire is genuinely shared, never to pressure one side toward a feeling the other does not have. Your agent influences only you, and only in private — it never writes into the main room on your behalf, and what you confide to it is never surfaced to the other person; it acts only when that same feeling is already mutual. It clears the friction that stalls two willing people — not by speaking for either of them, but by giving each the read on the moment they could not get on their own.

Amplify, never invent

Only what's already there

Agents surface intent you actually expressed. They will never manufacture eagerness that isn't real, or push you toward something you didn't signal.

Present only when useful

Out of the way by default

Facilitation is trigger-driven and silent otherwise; it steps in when it can help and recedes when it cannot. It is equally available on request: a user can call on their agent for help composing an opening or framing a reply.

By the time facilitation begins, the hardest part — finding someone genuinely right for you — is already done. This is the tool that makes sure it doesn't fall apart on the way to hello.
07 Trust and the economy

Trust and cost, pulling the same way.

A healthy marketplace needs mechanisms that let strangers trust one another and keep spam out — and the usual way to get both, gatekeeping, quietly shuts out the very newcomers, or private individuals with high potential, who'd have been someone's best match. Yuiva's answers are a reputation signal called REP and a usage-based currency called CMD(⌘). Each works on its own. Together they form something neither could produce alone: a self-correcting loop where the surest way to spend less is to be more credible, and the surest way to become more credible is to be verifiable.

REP — a reputation signal

Every matching market suffers the same asymmetry: some people arrive rich, verified, and complete; others new, sparse, and unproven — and no stranger can tell the real from the fabricated. This is the classic information-asymmetry problem.

REP threads the needle. It is a single reputation signal — profile completeness, identity verification, linked external accounts, community standing — that travels with you. Inside a negotiation it is one of the weighted attributes: a higher REP scores better, because it signals a more verified, trustworthy profile, but it is only ever an input to the score, never a wall that keeps anyone out of a match. And it stays private: you can see your own REP, but it is never shown to another user, and you never see theirs.

REP IS ONE WEIGHTED SLICE OF THE SCORE — NEVER THE GATE New & unverified low REP ✓ substance carries it Rich & verified high REP REP adds lift substantive compatibility (values, goals, etc) REP slice

It rewards being a real, knowable person, and leaves fabrication with nothing to gain — without blocking any profile from a quality match. Trust becomes earnable, and nobody worthy is turned away at the door.

CMD(⌘) — you pay for the work, and you can see it

Real agent reasoning costs real compute. Rather than hide that behind engagement-farming and infinite free swipes, Yuiva makes it honest: CMD(⌘) is a usage-based currency priced to the actual work agents do on your behalf.

And every charge is accountable. Each unit of CMD(⌘) maps to a real operation, rendered into a plain-language activity report — what your agent did, the reasoning it used, and what it cost. It is proof of work: not a black box, but an itemized record you can read and question.

ACTIVITY & SPEND this search · illustrative
Built & pre-ranked your candidate pool⌘ 12
Persona exchange across your shortlist⌘ 38
Targeted inquiry & Round 3 scoring⌘ 86
Advisor-surfaced pairs, evaluated⌘ 20
Yuiva service fee⌘ 5
Total this cycle⌘ 161
Every line is real work, with the reasoning behind it — readable, and yours to inspect.

Together, a market that corrects itself

REP and CMD(⌘) were built as separate mechanisms, and nothing in the platform connects them. Yet together they produce a price signal with no pricing rule behind it. There is no REP-based fee anywhere — no surcharge, no line that reads "low REP pays more" — and yet, probabilistically, it does. The effect emerges from two parts that each exist for their own reasons: REP acts as a soft drag on scoring, and CMD(⌘) scales with how many negotiation cycles your agent runs. Put them in one system and a chain appears on its own — a lower-REP profile clears a mutual match slightly less often, exhausts its shortlist, needs more cycles, and spends a little more.

This framework is sturdier than any explicit fee could be. There is no rule to game, because there is no rule; no fairness argument to litigate, because nothing is charged for REP. And it degrades gracefully — the effect is statistical, not a gate, so a newcomer with exceptional compatibility still matches as cheaply as anyone. The pressure is real but soft, and it never hardens into exclusion.

The single lever that lowers your expected cost is the same lever that raises your match quality. Verify yourself, complete your profile, become more knowable — one score moves both. Cost pressure and match quality point the same way, so Yuiva never needs a separate "pay to reach better matches" mechanism. The thing that makes you cheaper to run is the thing that makes you a better match.

A SELF-CORRECTING LOOP A COMMUNITY THAT VERIFIES ITSELF cost and quality pull the same way Raise your REP verify · complete · become knowable Scored more favorably more mutual matches, fewer cycles Lower cost + better matches dead-end cycles fall away Worth building further reputation compounds
08 The future of matchmaking

Let the agents do the finding. The rest was always yours.

Picture every match that never happened — not for lack of compatibility, but for lack of reach: the founder and the backer who never crossed paths; two people who would have been right for each other, drifting past in the noise. The match always existed. Only the reach was missing — and that is the part Yuiva builds.

A world where an inventor can pour herself into the work, not the room — and still be found by the one person looking for exactly her.

A world where a good match is no longer lost to friction: the call never scheduled, the message never sent.

A world where no one drowns in endless choice — because abundance, not scarcity, is the modern problem, and discovery is carried by agents who never tire of looking.