twin3 is the missing human layer of the agent economy—providing sovereign digital twins for people (xHuman.ID) and AI-native web identities for businesses (xAgent.ID) so that autonomous agents can discover, verify, understand, and contract with real human counterparties under cryptographic consent.

Autonomous AI agents can mathematically verify other agents via signed certificates and MCP tool schemas, but they cannot verify whether a real person is behind an inquiry or obtain informed human consent. As a result, agents fall back to Web2 methods—scraping stale contact lists and spamming unconsented inboxes—which breaks down as synthetic content floods the open web.

All seven foundational layers shipped in 2026—invocation (MCP), settlement (x402 Foundation), machine identity (Know-Your-Agent v1.0, FIDO), and discovery (merchant catalogs)—were built exclusively for corporate entities or machine-to-machine runtimes. Zero shipped layers provided a machine-readable address that an individual holds and an external counterparty can query under consent.

twin3 does not compete for narrow decentralized identity budgets. It re-plumbs the $50B+ global market currently spent on intent data, B2B data brokers, market research panels, and customer acquisition, alongside orchestrating a slice of the estimated $3–5 Trillion in annual agent-mediated commerce projected by McKinsey by 2030.

First, scraped data describes an absent subject who never agreed to the inquiry and is not compensated, providing zero incentive to maintain freshness. Second, behavioral traces are stale by the time they are purchased. Third, model-generated text increasingly pollutes public profiles, meaning heuristic inference fails precisely when demand for human signal peaks.

For AI agents to interact with, evaluate, or represent humans without invasive data surveillance, human characteristics, preferences, and permissions must be mapped into a high-dimensional, normalized mathematical vector (the 256D Twin Matrix) that machines can compute over deterministically.

In Web 2.0, users visited websites; in Web 3.0, wallets connected to decentralized contracts; in Web 4.0, services and external agents seek out the human's Personal Agent. Human beings participate continuously in the global agentic economy through their sovereign, 24/7 digital twin.

In an era of near-zero marginal cost for synthetic content, code generation, and computational reasoning, subjective taste, lived human experience, moral alignment, and physical-world verification are the only inputs AI models cannot hallucinate or self-generate.

twin3 is an infrastructure protocol connecting two product surfaces: xAgent.ID (B2B SaaS and Enterprise Agent) and xHuman.ID (consumer identity and personal agent), operating under an exchange model where enterprise demand funds consumer asset accumulation.

twin3 publishes its own verified enterprise agent card at https://xagent.id/twin3.ai, with machine-readable metadata and MCP endpoints available at https://xagent.id/twin3.ai.json.

The twin3 SDK is a unified cross-runtime protocol library comprising: L1 Discovery, L2 Trust/Identity, L3 Communication/Invocation, L4 Shield/Gating, L5 MatrixCore/Understanding, L6 Settlement/Matching, and L7 Governance/Audit.

twin3 Adopts public commons where the market has settled (L3 MCP/A2A, L7 standard audit trails); Builds infrastructure where required but non-proprietary (L1 discovery, L2 multi-rail identity, L6 matching); and Claims deep proprietary patent moats strictly where no person-side alternative exists (L4 Gating and L5 Understanding).

MCP is a rapidly adopted public standard with ~500M monthly SDK downloads in 2026. Competing at the raw invocation runtime creates needless adoption friction. twin3 implements MCP natively so any standard agent can interact with twin3 agents immediately without new protocol tooling.

L4 evaluates inbound machine requests against owner-defined value thresholds, schedules, and security boundaries. It features a five-stage prompt injection filter and attention pricing mechanism, refusing or pricing low-value automated solicitations before they reach human attention.

Instead of static strings or demographic checkboxes, L5 resolves the entity into a 256-dimensional normalized vector space, allowing mathematical distance calculations, semantic matching, and zero-knowledge attribute verification in sub-millisecond execution time.

L6 allows an autonomous agent to negotiate terms, prices, or deliverables on behalf of its human principal, but strictly within pre-signed geometric bounds (minimum price, maximum discount, delivery deadline) that the agent mathematically cannot exceed.

twin3 aligns with the x402 Foundation standard (Visa, Mastercard, Stripe, Google, AWS), returning standardized 402 Payment Required headers with payment options when an outside agent queries gated human attention or paid agent skills.

The core protocol library has 200 passing automated unit and integration tests and green CI across four language runtimes, validating canonical serialization, byte parity, challenge-response handshakes, and 256D vector arithmetic.

The enterprise inbox and personal inbox solve the same fundamental problem in reverse: one protects a business from spam while qualifying buyers; the other protects an individual while pricing attention. Using one shared protocol prevents divergent standards between supply and demand.

L7 uses hash-chained append-only audit receipts. The specific personal facts remain client-side or zero-knowledge committed; only the dated cryptographic receipt of consent, negotiation boundaries, and payment settlement is committed to the immutable log.

A 16x16 grid consisting of 256 cells. Each cell contains a 1-byte hexadecimal value from 00 (no signal / unverified) to FF (maximum verified depth), representing a standardized dimension of identity or capability.

Four quadrants of 64 dimensions each: 1. Physical Me (sleep, stamina, sensory comfort, location radius); 2. Digital Me (skills, credentials, learning pace, wallet history); 3. Social Me (languages, community roles, communication style); 4. Spiritual Me (moral boundaries, aesthetic taste, risk attitude, fairness index).

The Enterprise Twin Matrix allocates its 256 cells across four distinct commercial quadrants: Physical/Legal (80 cells: corporate registration, physical sites, compliance), Digital/Agentic (100 cells: domains, MCP tool schemas, API latency, security), Social/External (40 cells: citations, partner attestations), and Policy/Vibe (36 cells: privacy, AI safety policies, human approval thresholds).

Values are initialized through connected cryptographic accounts, onboarding calibration tasks, and domain verifications. They update dynamically as the entity completes verified tasks, receives signed counterparty evaluations, and renews decaying leases.

The inspecting agent queries the user's Personal Agent with specific requirements. The Personal Agent uses zero-knowledge selective disclosure proofs (US Pat. App. 64/142,929) to prove that relevant dimensions satisfy threshold constraints without revealing exact scores or unconsented quadrants.

JEV acts as a deterministic System One evaluator. Instead of passing the entire matrix into a heavy LLM prompt, JEV evaluates permitted matrix slices against typed rules to return instant yes/no classifications, confidence probabilities, or match scores in milliseconds.

The Human Twin Matrix is anchored as an ERC-4671 Soulbound Token on BNB Chain under verified contract address 0xE3ec133e29adDfbBA26a412c38ed5De37195156f.

Over 143,619 credentials have been minted on BNB Chain (publicly cited as 140,000+), ranking twin3 #7 globally among all SBT projects and #2 on BNB Chain behind Binance BAB.

No. twin3's disciplined disclosure standard strictly maintains that 140,000+ is a count of issued credentials proving account control and a maintained record; it does not prove biological uniqueness or personhood, and twin3 never presents it as such.

The protocol was published in the peer-reviewed IEEE Access journal in 2024: 'twin3: Pluralistic Personal Digital Twins via Blockchain' (Wen & Lin), establishing formal mathematical proofs for decentralized multidimensional twin representation.

xHuman.ID is the sovereign personal identity layer of twin3, giving each individual a machine-readable address (xhuman.id/{handle}) that declares verifiable credentials, an explicit list of what it does not prove, and owner-defined contact policies.

It is a cryptographic assurance mechanism where an identity proof is issued with a finite epoch. Unless actively renewed through continuous account activity, human challenge-response, or multi-party refresh, the assurance score decays over time, neutralizing zombie credentials and sold accounts.

If an identity credential never expires (like a standard NFT or static iris registration), private keys can be sold, stolen, or harvested into industrial account farms that rent out 'verified human' status to automated bots without detection.

Outside agents never communicate with the human directly. They interact with the Personal Agent, which screens incoming requests against the owner's sovereign usage policy (Allow / Ask / Deny) and minimum attention price before alerting the user.

Users manage their Personal Agent and Twin Matrix through @twin3clawbot on Telegram, providing an intuitive interface to configure 11 modular capability slots and inspect incoming agent inquiries.

They are governed execution slots where users equip capabilities (e.g., Job Opportunity Matcher, Dating/Social Filter, Brand Taste Panelist, Verification Witness) that execute autonomously strictly under owner-delegated rules.

Free ($0/mo): Create twin and Twin Matrix with basic identity reads; Plus ($4.99/mo, primary paid plan): AIInbox opportunity matching for jobs and relationships; Creator ($19/mo): Fan cards and community monetization.

twin3 uses a conservative 1% paid conversion baseline: targeting 10,000 paying Plus/Creator members from a 1,000,000 active card base by mid-2027.

A representative live public card is xhuman.id/mreagle (Card No. 8473), demonstrating public biographical presentation, Twin Matrix visual hashes, and machine-readable JSON policies.

Yes. Sovereign usage policies are bounded by time epochs (e.g., 30-day lease) and cryptographically revocable at any moment by the owner through on-chain key rotation or policy update.

xAgent.ID is the business-facing identity and agent infrastructure that turns any public domain into an AI-native company website and verified Enterprise Agent that external buyer agents can discover, trust, and purchase from.

SMEs lack multi-million dollar AI engineering teams to build custom agent infrastructure. xAgent.ID gives them an out-of-the-box machine-readable presence, AIInbox, and automated quoting agent for as little as $12.99/month, enabling them to win orders from autonomous buyer agents.

The front surface is an aesthetic, human-readable responsive web profile; the back surface is a structured, machine-parsable interface serving agent-card.json, OpenAPI/MCP tool definitions, rate limits, and cryptographic passport signatures.

xAgent.ID binds enterprise identities to Base Chain via the ERC-8004 Identity Registry contract (0x8004A169FB4a3325136EB29fA0ceB6D2e539a432).

Identity ($12.99/mo): Verified domain passport and machine-readable agent card; Monitor ($29.99/mo): AI crawler analytics and visit audits; Agentic ($99/mo): Live AIInbox, prompt defense, and automated RFQ parsing; Operate ($199/mo): Bounded agent orchestration and multi-domain management.

1. AIInbox: Receives A2A inquiries, RFQs, and orders in one screened queue; 2. Multilingual Support: Qualifies buyers across global languages; 3. Company ASO: Agent Search Optimization to ensure top discovery in LLM searches; 4. Agent Listings: Syncs business presence across global machine directories.

Inbound agent payloads pass through a multi-stage sanitizer that strips executable formatting, validates counterparty domain signatures, enforces context bounds, and isolates external input from core business execution logic.

The Enterprise Agent receives an RFQ via MCP/A2A, verifies buyer capability, computes terms within the owner-approved authority envelope (US Pat. App. 64/142,916), generates a signed quote, and captures payment via HTTP 402 upon human or policy sign-off.

The package @twin3-ai/agent-id is publicly published on npm at version 0.3.49 across 51 releases, allowing developer teams to evaluate and integrate enterprise agent identity locally.

xagent.id/binance.com is a demonstration card generated from public corporate records to illustrate how buyer agents read capabilities; twin3 explicitly labels it as unclaimed to maintain strict evidentiary standards.

twin3 holds nine U.S. provisional patent applications filed with the United States Patent and Trademark Office (USPTO). All rights are assigned to TWIN3 AI LIMITED (Hong Kong), with Ming-Hui Wen as the sole inventor.

Applications 001 (64/122,823) and 008 (64/122,848) have priority dates of 30 July 2026. The remaining seven applications (002–007 and 009) have priority dates of 28 August 2026.

Application 001 covers a 'Decaying Personhood-Assurance Lease' (Key Drawings: FIG. 5 & 11): a protocol where proof of live personhood naturally decays and must be continuously re-earned through non-transferable, rate-bounded signals. Account events (revocation, unbinding, key recovery) atomically increment a continuity epoch, immediately terminating downstream machine credentials without inheritance.

Application 002 covers a 'Sovereign Self-Evolving Human Representation' (Key Drawing: FIG. 4): technically separating AI inference from authority to represent statements as the human subject's words. Inferences require subject sovereign publication policies and automatically demote or withdraw when source evidence lapses.

Application 003 covers 'Value-Based Routing of Machine-Initiated Opportunities' (Key Drawing: FIG. 6): treating human attention as a policy-controlled resource, algorithmically evaluating inbound machine requests against a multidimensional expected-value vs. cost tensor (risk, time, privacy, attention).

Application 004 covers 'Autonomous Agent-To-Agent Negotiation Under Authority Envelopes' (Key Drawing: FIG. 6): enforcing geometric parameter constraints on agent contracting (price, SLA, terms). An agent cannot issue binding commitments outside its pre-authorized envelope without counter-signed human approval proofs.

Application 005 (U.S. 64/142,921) covers Multi-Rail Identity Reconciliation without trusting caller assertions; Application 006 (U.S. 64/142,927) covers Runtime Custody Migration & Continuity preserving stable identity across provider and on-premise execution; and Application 007 (U.S. 64/142,929) covers Selective Disclosure using private commitment ledgers and purpose-bound minimum proofs.

Application 008 (U.S. 64/122,848) covers Credential-Less Control-Plane Modification via three-key separation and derived-expectation public re-verification; Application 009 (U.S. 64/142,930) covers Generative AI Visibility Closed Loops that measure answer citations and execute bounded remediation work orders.

No. All nine filings are strictly labeled as 'Patent Pending' (U.S. provisional patent applications filed). twin3 adheres to institutional legal compliance, never refers to them as issued patents, and explicitly clarifies that patent protection covers underlying cryptographic and architectural mechanisms rather than visual profile cards.

1. Intellectual Property (9 patent applications); 2. Two-Sided Network (xHuman.ID user population and xAgent.ID domain base); 3. Protocol Standard (interoperable gating and envelope contracts); 4. Intelligence Depth (accumulated 256D behavioral history in the Twin Matrix that cannot be scraped or bought).

Loop 1: Enterprise agent readiness subscriptions (selling now); Loop 2: Personal paid inbox attention fees (in build); Loop 3: Human-to-human matching fees (in build); Loop 4: Physical-world task economy exchange (specified).

For direct transactions: 80% to the human contributor, 20% to the twin3 platform. For referred transactions: 70% to the human, 20% to the platform, and 10% to the referrer perpetually. For enterprise contracts: 70% human, 25% platform, 5% channel partner.

Web2 platforms (Google, Meta, Uber, Upwork) depend on taking 70% to 90% of value from contributors to support massive corporate overheads. They cannot match twin3's 70% payout without dismantling their own core business models and destroying their share prices.

Selling API calls is a race to the bottom with zero network effects. Building a consented exchange between enterprise buyer agents and verified personal twins creates two-sided compounding: more credential holders make any query more resolvable, and more transactions make each twin's history more valuable.

Enterprises have existing commercial budgets to be discovered and win orders, generating immediate SaaS cash flow. This revenue funds protocol engineering while the scarce consumer asset (maintained personal credentials) accumulates organically without venture-subsidized user acquisition burn.

The $twin3 token provides the low-friction cross-border settlement rail for micro-transactions across 122 countries (where credit card minimums make $1 payouts impossible) and powers the credential maintenance sink required to refresh decaying leases.

The primary sink is credential maintenance: holders spend tokens to keep their decaying leases resolving. Secondary sinks include query settlement and higher-tier exposure. The faucet is a contribution index rewarding verification depth and completed tasks—never issued for passive holding.

TGE is gated strictly on achieving credential base milestones (e.g., crossing 500,000 active resolving twins) rather than an arbitrary calendar date, ensuring deep economic utility before liquidity.

twin3 is pre-revenue. Tiered domain pricing is published, Stripe billing infrastructure is built, and initial paid customer conversions are underway through the GTMC channel.

Blended gross margin is modeled at 75%, driven by the high gross margins of software subscriptions (85%+) offset by cloud infrastructure and multi-chain gas relay costs.

Worldcoin provides a binary 1D proof: 'Is this an unbanned human eye? Yes/No.' It tells an agent nothing about the person's skills, taste, languages, or availability, and provides no gating mechanism. twin3 provides full 3D computable depth (256D Twin Matrix) plus sovereign attention gating and dynamic decaying leases.

Bittensor builds decentralized subnets to incentivize machine intelligence (LLM training, compute, inference). twin3 does not compete in compute subnets; it builds the complementary human layer—supplying verified human judgment, taste alignment, and authentic consent that compute networks cannot self-generate.

Grass and Vana monetize passive web scraping or residential IP bandwidth, collecting raw unstructured web text. twin3 captures structured, active, sovereign human judgment, taste, and consent, compensating individuals directly under a 70% revenue share.

Galxe and Gitcoin Passport aggregate web3 badges into a static anti-sybil score. They do not provide an agent-to-agent communication layer, cannot be queried via MCP by enterprise AI agents, and do not feature dynamic decaying leases or attention-pricing inboxes.

Brokers sell unconsented, static, scraped copies of people who receive zero compensation. twin3 introduces the person's own agent into the transaction as a paid counterparty, delivering fresh on-demand verification, complete legal consent, and zero synthetic noise.

Card networks excel at financial settlement (e.g., the x402 Foundation and AgentCard), but they settle transactions between cards and merchants. They do not own or maintain multidimensional behavioral vectors or personal attention firewalls for individuals.

Hyperscalers build model intelligence and agent runtimes. A human identity and consent layer requires neutrality across all models and frameworks; an identity tied exclusively to Google or OpenAI would be rejected by the broader ecosystem.

A competitor can fork open-source code, but they cannot fork: 1. The nine pending patent claims covering decaying leases, authority envelopes, and attention routing; 2. The 140,000+ credential network on BNB Chain; 3. The 1,000+ SME channel via GTMC; 4. The historical behavioral depth accumulated within the Twin Matrix.

Most marketplaces fail because they launch both sides cold. twin3 solves this by monetizing the enterprise side immediately through existing SME service partnerships while accumulating consumer credentials organically via warm-start communities and perpetual referral incentives.

twin3 is privacy-by-design: raw personal data is never written to public blockchains. The chain stores only zero-knowledge commitments and normalized vector hashes. Because the user holds their private keys and grants decaying leases, full sovereign right-to-revoke is maintained by construction.

1. GTMC: B2B web services provider with 1,000+ Taiwanese manufacturing clients; 2. Empowerfeel / StockFeel Group: Financial marketing platform with 1M+ consumer users; 3. Coral AI: Enterprise AI and semiconductor algorithm partner.

twin3 has closed $400,000 in strategic funding at a $10M post-money cap: $200,000 from Empowerfeel, $100,000 from GTMC, and $100,000 from Coral AI.

GTMC is co-developing the Enterprise Agent with twin3 and introducing it to their existing book of 1,000+ manufacturing clients as an AI-native export upgrade, converting an established web budget line directly into xAgent.ID subscriptions.

StockFeel connects over 1,000,000 financial consumers, providing a direct top-of-funnel pipeline for individuals to create digital twins and participate in brand research tasks.

Taiwan possesses an export-oriented, high-density manufacturing base that conducts cross-border commerce in foreign languages—the exact scenario where machine-readable agent profiles outperform traditional search—within a legible, saturable geography.

Founder Ming Wen previously built a consumer messaging bot that scaled to 1.2 million users with zero paid marketing, and an online community platform that reached 300,000+ active members organically.

1. Champion (US$5,000 1st prize) at the 2026 Trustworthy AI Hackathon (Taipei, 50 teams); 2. Top 4 winner at the Web3 Festival RWA Demo Day (HKU, 100+ applicants); 3. Demo Day Best Presentation at the RWA Hackathon Taiwan (National Tsing Hua University).

Over 10,000 active community members on the project's official server and 5,000+ followers on X, built entirely through organic distribution without paid ads.

twin3 offers a free, reproducible agent-readiness score for any public website. The audit exposes discoverability gaps for AI crawlers, acting as the primary inbound lead magnet for paid xAgent.ID subscriptions.

A Google Cloud Marketplace listing is currently in review, which will allow enterprise procurement teams to purchase xAgent.ID subscriptions against existing committed cloud spend.

twin3 is raising $3,000,000 at a $15,000,000 post-money valuation via an Equity-led SAFE, accompanied by a $twin3 Token Warrant.

Founder Ming Wen retains 68.0% (down from 97.96% initial); Seed investors hold 20.0%; strategic investors (Empowerfeel, GTMC, Coral AI) hold 5.6%; and an additional strategic tranche holds 6.4% (before new ESOP pool allocation).

FY2026: $0.02M (planning baseline) -> FY2027: $1.75M -> FY2028: $8.46M -> FY2029: $24.72M.

xAgent.ID subscriptions: $19.50M (45,000 paid domains); xHuman.ID memberships: $1.50M (30,000 paying Plus/Creator members); Agent order fees: $3.00M (1% take on completed outcomes); On-prem enterprise plans: $0.72M.

twin3 operates on an agentic cost structure: development, testing, customer onboarding, and tier-1 support are handled by internal AI agents. Opex grows from $1.00M in FY27 to only $2.00M in FY29, dropping from 57% to 8% of revenue.

At an 8x multiple on FY2029 subscription ARR ($29.88M), the implied enterprise valuation is $239.0M, yielding a 9.2x return on the $3M Seed investment (assuming 11.52% ownership post-subsequent Series A/B dilution).

1. First external third-party agent completing an A2A handshake through AIInbox; 2. Conversion of initial paid enterprise domains via GTMC; 3. Cloud marketplace listing live; 4. Beta release of consented selective dimension sharing on xHuman.ID; 5. U.S. patent conversion decisions in Q1 2027.

Consumer adoption of portable credentials: if individuals remain content being understood and siloed by dominant platforms without seeking sovereign portable representation, the consumer exchange asset will accumulate more slowly, confining twin3 to a conventional B2B software business.

Hyperscalers can build identity verification rapidly, but they cannot replicate the accumulated multidimensional behavioral depth of the Twin Matrix or position themselves as a neutral party trusted by individuals to refuse requests across competing platforms.

Inspect the live verified Enterprise Agent card at https://xagent.id/twin3.ai and machine JSON at https://xagent.id/twin3.ai.json; verify the smart contract on BNB Chain (0xE3ec133e29adDfbBA26a412c38ed5De37195156f); audit the published package @twin3-ai/agent-id on npm; and review the peer-reviewed IEEE Access paper (DOI: 10.1109/ACCESS.2024.10734204).