DUALITY AI · INVESTOR THESIS

We do not need to outspend the frontier labs to build a serious AI company.

We need to win where size becomes friction.

Duality is building a local-first, privacy-aware AI platform and distributed compute network beginning in Brazil — turning useful workflows, trusted hardware, and measurable execution into a path toward serious model capability.

Local, Trusted, and Community compute layers converging through a routing coreLOCALTRUSTEDCOMMUNITY

The Central Proposition

Useful AI work becomes infrastructure.

Duality begins with real, narrow workflows rather than a generic chatbot. Each successfully completed task teaches the system something about routing, cost, and quality that a broader competitor has little reason to learn first.

THE INITIAL WEDGE

  • Private document intelligence
  • Portuguese and English business workflows
  • Summarization and structured extraction
  • Batch document processing
  • Local assistants for sensitive material
  • Trusted nodes for higher-value workloads

Architecture

The three-tier compute architecture

LOCAL

Runs on the user's own machine.

BEST FOR

  • Private documents
  • Offline use
  • Sensitive material
  • Personal context
  • Low-latency local tasks

TRUSTED

Runs on vetted infrastructure or approved operators.

BEST FOR

  • Enterprise workloads
  • Stronger models
  • Larger contexts
  • Controlled data processing
  • Higher-reliability jobs

COMMUNITY

Runs approved, non-sensitive batch work across participating machines.

BEST FOR

  • Public data
  • Synthetic workloads
  • Evaluation jobs
  • Large non-sensitive batches
  • Cost-sensitive asynchronous work

Privacy is not a marketing promise added after deployment. It is a routing decision made before execution.

Positioning

The giants are stronger. They are not optimized for every battle.

FRONTIER LABORATORIES

Optimize for broad global capability

DUALITY

Begins with narrow, painful workflows

FRONTIER LABORATORIES

Depend on enormous centralized infrastructure

DUALITY

Combines local, trusted, and distributed capacity

FRONTIER LABORATORIES

Must serve global averages

DUALITY

Can obsess over Brazil and Portuguese

FRONTIER LABORATORIES

Compete primarily through model scale

DUALITY

Competes first through architecture, workflow, trust, and cost

FRONTIER LABORATORIES

Carry major organizational and infrastructure inertia

DUALITY

Can iterate quickly with a focused team

FRONTIER LABORATORIES

Treat regional specialization as one priority among thousands

DUALITY

Can make regional specialization the company's foundation

Structural Advantages

Where the leverage actually comes from

BRAZIL AS AN OPERATING ADVANTAGE

A major population and economy with genuine Portuguese-language needs and underserved regional workflows.

  • Lower operating costs than major American technology centers
  • A growing technical community
  • Demand for national and private AI capability
  • The ability to build for Brazil rather than merely translate into Portuguese

LOCAL-FIRST ARCHITECTURE

Execution that starts on the user's own hardware changes the privacy and cost equation entirely.

  • Privacy by default, not by policy
  • Resilience during connectivity loss
  • Cost control for high-volume tasks
  • User ownership of sensitive material
  • Reduced dependence on one central provider

DISTRIBUTED COMPUTE SUPPLY

Idle machines become useful capacity — but only once verification, metering, and trust exist first.

  • Reputation scoring across nodes
  • Hardware benchmarking before workload assignment
  • Reliability tracking over time
  • Trusted tiers with restricted workload classes

WORKFLOW AND EVALUATION MOAT

The durable advantage is not raw compute. It is what is learned by running real workflows repeatedly.

  • Routing logic tuned by observed outcomes
  • Customer-specific evaluation sets
  • Portuguese-language quality benchmarks
  • Verified node histories
  • Cost-per-successful-task knowledge

CAPITAL DISCIPLINE

The first objective is not to possess the most compute. It is to understand exactly where compute creates durable value.

  • Prove workflow value before frontier-scale spend
  • Measure cost per verified task from day one
  • Expand compute commitments only against evidence

The Flywheel

A loop that compounds with use

The Climb

We earn the right to attempt the frontier.

  1. Stage I

    PRODUCT PROOF

    NOW
    • Local model execution
    • Document intelligence
    • Portuguese/English workflows
    • Benchmark harness
    • First users
  2. Stage II

    NETWORK PROOF

    NEXT
    • Miner client
    • Scheduler
    • Model registry
    • Hidden audits
    • Reputation scoring
    • Verified token accounting
  3. Stage III

    COMMERCIAL PROOF

    LATER
    • Paid pilots
    • Trusted compute tier
    • Repeatable task economics
    • Enterprise privacy controls
    • Measurable retention
  4. Stage IV

    PROPRIETARY BEHAVIOR

    LATER
    • Retrieval systems
    • Fine-tuned adapters
    • Portuguese evaluation suites
    • Specialized routing
    • Domain-specific models
  5. Stage V

    SERIOUS MODEL RESEARCH

    LONG-TERM
    • Owned and partner compute
    • Distributed training research
    • Proprietary datasets
    • Research and evaluation team
    • Foundation-model work where strategically justified

Compute Contributors

Aligned ownership, not speculative investing

COMMUNITY MINERS

Contribute spare capacity to non-sensitive, asynchronous batch workloads. Compensated through credits or cash economics.

VERIFIED OPERATORS

Established reliability and benchmark history. Eligible for a wider range of workload classes.

TRUSTED COMPUTE OPERATORS

Vetted infrastructure approved for higher-value and privacy-sensitive workloads.

STRATEGIC INFRASTRUCTURE PARTNERS

Long-term, large-scale contributors. May eventually be eligible for a formal contributor-equity program for exceptional long-term infrastructure partners, subject to KYC, service requirements, board approval, securities law, tax review, and individual grant documentation.

Duality may eventually create a formal contributor-equity program for exceptional long-term infrastructure partners, subject to KYC, service requirements, board approval, securities law, tax review, and individual grant documentation. Normal compute contribution is compensated through credits or cash economics — nothing here implies a guaranteed equity outcome.

Diligence

The questions that decide whether we win.

Can useful AI jobs be reliably routed across heterogeneous machines?

Can the system verify that the correct model performed the requested work?

What is the true cost per useful, verified task?

Which workflows are cheaper or more private than hosted-only alternatives?

Will compute contributors remain active?

Will users repeatedly pay for the selected workflows?

Can Portuguese quality become demonstrably better in the target domains?

Can trusted infrastructure satisfy enterprise privacy requirements?

Can the network maintain quality as it scales?

Duality’s early value will be measured by evidence, not narrative.

Operating Scorecard

What we intend to measure — not what we claim today

No production metrics exist yet. The items below are the measurement framework Duality intends to track and report against as the product and network mature.

Useful verified tasksMeasurement framework — to be validated
Cost per successful taskTarget metric — to be validated
Active compute nodesMeasurement framework — to be validated
Trusted-node availabilityTarget metric — to be validated
Audit pass rateMeasurement framework — to be validated
Job completion rateTarget metric — to be validated
Median latencyMeasurement framework — to be validated
Repeat usageTarget metric — to be validated
Paying customersMeasurement framework — to be validated
Gross margin by workflowTarget metric — to be validated
Portuguese-language evaluation performanceMeasurement framework — to be validated
Privacy-routing failuresTarget metric — to be validated
Contributor retentionMeasurement framework — to be validated

Candor

Risks, and the current strategic response

RISK

Consumer nodes may be unreliable.

RESPONSE

Begin with asynchronous whole-job routing, reputation scoring, duplicate sampling, trusted tiers, and cloud fallback rather than attempting fragile multi-node interactive inference immediately.

RISK

Security and privacy failures would be severe and reputation-ending.

RESPONSE

Treat privacy as a routing decision made before execution, not a policy applied after the fact. Sensitive material defaults to local or trusted tiers only.

RISK

Underlying open-weight models could commoditize any single point of differentiation.

RESPONSE

Build the moat in workflow design, evaluation data, and routing — not in any one model — so capability gains upstream benefit Duality's stack directly.

RISK

Customer acquisition in a crowded AI-tools market is difficult.

RESPONSE

Start with narrow, painful, underserved workflows in Portuguese-first markets rather than competing head-on for generic assistant usage.

RISK

Verification of distributed compute work carries real overhead cost.

RESPONSE

Use duplicate sampling, hidden audit jobs, and reputation-weighted trust rather than verifying every task at full cost.

RISK

Regulatory treatment of distributed compute and any contributor-equity program is uncertain.

RESPONSE

Treat any equity-like compensation as a formal, counsel-reviewed program gated by KYC, service requirements, and securities law — not an ad-hoc token mechanic.

RISK

Supply and demand for compute and workflows may not balance.

RESPONSE

Grow the contributor network deliberately in step with paid workload volume rather than recruiting capacity ahead of demonstrated demand.

RISK

Later-stage model research requires capital Duality does not have on day one.

RESPONSE

Sequence the roadmap so each stage is funded by evidence generated in the prior stage, deferring frontier-scale spend until it is justified.

Timing

Why now

Increasingly capable open-weight models

Increasingly effective local inference

Widespread consumer and professional GPUs

Growing business demand for AI workflows

Rising concern about privacy and centralization

Growing importance of regional and language specialization

Better serving, quantization, retrieval, and fine-tuning techniques

Duality does not win by pretending capital does not matter.

It wins by using capital more intelligently. The company begins with narrow utility, converts idle hardware into verified capacity, makes privacy an architectural choice, and builds in a market the frontier labs cannot afford to study with the same intensity.

If the first wedge works, Duality is not merely another interface over someone else’s model. It becomes a compute network, a workflow company, an evaluation company, a trust layer, and eventually a model company. That is the asymmetric possibility.

REQUEST INVESTOR MATERIALSTECHNICAL THESIS — COMING SOON

This page is provided for informational purposes only and does not constitute an offer to sell, or a solicitation of an offer to buy, any security. Any potential investment opportunity would be made only through appropriate legal documentation and in compliance with applicable law.