Neural Arc Inc. · Seed Round · USD 10,000,000

The Sovereign
Intelligence Stack

A ten million dollar seed round to own the layer between every intelligent application and every model that runs it.

← →  arrow keys, or swipe

I  ·  The ProblemThe rented default

The application is yours. The intelligence is rented.

Software is being rebuilt with intelligence inside it, yet the computation underneath belongs to a handful of frontier laboratories. Every product built this way inherits three things it never chose.

PricingSomebody else sets it, and changes it without negotiation.
DeprecationA retired model becomes an unplanned engineering quarter.
Data boundaryRegulated workloads cannot cross it. Yours do, daily.
II  ·  The Engine  ·  ModelBeatmodelbeat.ai · Live

Own the engine, not the invoice.

ModelBeat is orchestration intelligence, not a marketplace. It reads a request, splits it, routes each part to the model that answers it best, and returns one answer with context retained.

Fig. 1One request, many models, one answer
RequestRead, understood, decomposed
↓
Part A↓Open WeightHandpicked by Elytra
Part B↓Your Own ModelBrought in, or built inside
Part C↓Frontier, GatedOnly when nothing else clears the bar
↓
One AnswerContext retained · every token, route, and cost logged
On premises, your own cloud, hybrid, or managed: nothing leaves the perimeter.
III  ·  The Moat  ·  ElytraPublished research

Benchmark the world, then own the cluster.

Elytra, our own evaluation protocol, benchmarks open-weight models continuously on quality, latency, and cost, and handpicks the cluster. Customers bring or build their own domain models alongside it. The method is set out in our own research: a position paper and a technical report.

37 to 43%lower token cost, measured in internal testing
450+models orchestrated across the routed field
One keyserves every application in the business
Every tokentraced to a person, a task, and a route
IV  ·  The FieldCompetitive position

Measured against the field.

Fig. 2Cost per identical workload, indexed
A single frontier contract100
ModelBeat, routed57 to 63

Saving of 37 to 43 percent, measured in internal testing.

Indexed to a single frontier contract at 100. The routed figure reflects the cheapest model that clears the quality bar on every call.
Fig. 3Orchestration against the field
ModelBeatRoutersMarketplacesFrontier direct
Many models behind one key●●●○
One request, decomposed and routed●◑○○
Context retained across models●○○○
Own benchmark protocol, refreshed●◑○○
Bring or build your own models●○○◑
On premises, per-token audit●○○○
● Full  ·  ◑ Partial  ·  ○ Absent  ·  Category-level view of the field.
V  ·  The StackInside your perimeter

One perimeter, four layers.

04 · The SurfaceArgus Speak, and work happens. Voice and agents that see the screen and act on it. Sept 2026
03 · The WorkspaceHelium AI Where the team works. Shared memory, permissions, and a view of every prompt and cost. Live
02 · The KnowledgeTensaris Your own data, made answerable. Every reply grounded in what the company knows. Live
01 · The EngineModelBeat Reads the request, routes each part, returns one answer. Every other layer stands on it. Live · Core

The workspace wins the user. The engine wins the infrastructure. A model can retire without the layer above it noticing.

VI  ·  TractionSince the mid-May 2026 launch

Already live. Already selling.

80,000+users since the mid-May 2026 launch
$2MARR contracted, revenue from October 2026
$11Mactive sales pipeline
$30Mfurther pipeline in late-stage NDA discussion

Late-stage NDA discussions with a Big Four firm, a large mining group, and a European bank sit outside the active pipeline figure.

VII  ·  The ChoiceRented, or owned

Retire the single-model default.

With the Neural Arc stack

One request, many models, one answer.
Thirty-seven to forty-three percent lower token cost.
Sovereign by architecture. Nothing leaves.
Every decision auditable, to the token.

The frontier default

One vendor sets the terms.
Spend that nobody can attribute.
Data crosses a boundary you do not control.
A retired model becomes an unplanned quarter.
VIII  ·  The FoundersNewark, Delaware · Pune, India

Founders who have done this before.

Aniket Tapre Aniket Tapre Founder and CEO

A serial entrepreneur with five exits across different domains and a proven record of building and selling technology companies. A product and technology founder who conceived the sovereign AI thesis early and now leads a team of thirty-five across the US, the UK, and India.

Sheetal Patole Sheetal Patole Co-Founder

A career at the top of the very enterprises Neural Arc sells to, with senior roles at Barclays, Rio Tinto, PwC, Macquarie Bank, and Deloitte. She knows exactly how large regulated organizations evaluate, buy, and govern technology, because she has run it from the inside.

IX  ·  The RoundSeed · 2026

Fund the layer, not the rent.

Total Raise$10,000,000Seed, at approximately USD 100,000,000 post-money
Minimum Ticket$300,000Per investor, USD
Equity releasedNo more than ten percent
Target returnEighteen to nineteen percent annualised
LiquidityOption to exit at the Series A
TeamThirty-five people across Delaware and Pune

Target return is a management target, not a guarantee. Private investments carry risk, including the loss of capital.

Rent intelligence, or direct it.

The application layer is crowded. The layer underneath it is not.

Products modelbeat.ai he2.ai
Company Neural Arc Inc. neuralarc.ai
Contact aniket@neuralarc.ai Aniket Tapre · Founder and CEO

Newark, Delaware · Pune, India  ·  Confidential · Prepared for prospective investors · 2026

01 · 10