Hiyaku · The Enterprise Autonomy Solution · Architecture

Every enterprise runs on fragmented systems. We turn them into one living model.

Today, most companies use automation to optimize and execute functional tasks. Hiyaku transforms how the enterprise operates by managing outcomes.

Hiyaku connects the systems, data, processes and decisions that run your business into one living model, an ontology. It uses that model to reinvent end-to-end processes and enables governed agents to pursue objectives and manage outcomes. Below, watch fragmented parts become one connected model, one ontology layer at a time.

Your systems. Your data. Your IP. Our solution.

Live theme

Kintsugi Parthenon

Drag to walk around

This browser could not start WebGL, so the 3D stage is off. The layer legend and every section below still apply.

Our approach

Keep what works about ontologies. Remove what keeps them out of reach.

The core insight is sound. Most of the limits come from who ontology platforms have been built for: large enterprises and governments with long contracts. Hiyaku changes the defaults, not the idea.

Objects, not tablesPeople work with Customers, Orders and Shipments. The ontology maps every source table onto those nouns.
Actions write backA decision made in an app changes the source systems, with rules and an audit trail, instead of ending in a spreadsheet export.
Lineage everywhereEvery dataset knows which inputs and which code produced it, so any number can be traced to its origin.
Apps on one modelDashboards, workflows and AI all read the same ontology, so they agree with each other.
QuestionTypical enterprise platformsHiyaku
Who can adopt itLarge enterprises and governments, often with vendor engineers deployed on site for months.One business unit starts from a domain template in days. The same core scales to the whole enterprise without a re-platform.
Where the ontology livesInside the platform, edited through its tools.An open, versioned spec in git. You can diff it, review it in a pull request, and export it to RDF, JSON Schema or SQL.
Where data is storedPlatform-managed storage and formats.Open table formats (Apache Iceberg on Parquet). Your data outlives any vendor, including us.
Where it runsVendor cloud or dedicated deployments.Your cloud, your data centre, or an edge site with patchy connectivity. Queries travel to data that legally cannot move.
How AI takes partAI features layered on top of the ontology.Agents are ontology users. They hold roles, call the same typed actions as people, and every step lands in the same audit log.
How permissions behaveStrong controls that specialists configure.Policies are attached to data and travel through every join and transform automatically, so a derived table can never be more open than its strictest source.

The complexity we are solving

Eight problems that look simple until you meet them at scale

None of these is solved by a better dashboard. Each one grows faster than the organisation does, which is why most enterprises end up with dozens of systems that disagree about basic facts. Each one is tagged with the step of Hiyaku's approach that solves it: Connect, Understand, Reinvent or Act.

N(N−1) ÷ 2

Connect

Integration sprawl

780possible point-to-point pipes between 40 systems. Through one shared ontology it is 40 mappings.

Each new system has to talk to every old one. A hub model makes the cost grow linearly instead of quadratically.

n² ÷ 2 → n·b ÷ 2

Connect

Who is who

1.6 yearsto compare 10 million records pairwise at a million checks a second. Blocking cuts it to about 8 minutes.

Is "Ravi K., 12 MG Road" the same customer as "R. Kumar, 12 Mahatma Gandhi Rd"? Entity resolution decides, and sends uncertain pairs to a person.

fact × valid time × recorded time

Understand

Two clocks per fact

2 axes"What did we believe on 3 March about stock on 1 February?" needs both.

Corrections are new rows, never overwrites. Audits, recalls and court cases depend on reconstructing what was known at the time.

policy(out) = strictest(policy(in₁…inₖ))

Understand

Permissions that follow data

6 sourcesbehind one dashboard means six rule sets. Miss one and a warehouse temp can see payroll.

Access is by role and by purpose (the DPDP Act 2023 and GDPR both care why you read data). Rules must propagate through every join automatically.

2ᵐ partial states

Act

Writing back safely

32ways a "reroute shipment" action touching 5 systems can half-succeed.

Actions run as sagas with compensating steps and idempotency keys, so a retry never double-books a bed.

column → transform → decision

Understand

Lineage you can trust

100sof transforms can sit under a single KPI on a board pack.

Column-level lineage lets you trace one bad source row forward to every report and decision it touched, then fix all of them.

agent ⊆ role ∩ actions ∩ audit

Reinvent

AI that can act

0raw SQL access for agents. They call typed actions or nothing.

An LLM with database credentials is a liability. One that proposes typed actions, checked against the same permissions as staff and replayable later, is a colleague.

move compute, not data

Connect

Sovereignty

Stays putCustomer data that may not leave a country is queried where it lives.

Federated queries push filters and aggregates down to each site and bring back only the permitted answer. Edge nodes keep working offline.

Try the arithmetic

Integration sprawl
Point-to-point pipes
780
Mappings to one ontology
40
Entity resolution
Compare every pair
1.6 years
With blocking (groups of 100)
8.3 min

Timings assume one million comparisons per second on one core. Real systems parallelise, but the ratio between the two columns stays the same.

The ontology, as a temple

Six layers, each resting on the one below

A Greek temple is a good model because its parts have jobs. Columns stand alone; the entablature only means something once it rests across two columns. Object types and link types work the same way.

Elevation of a Greek temple with each part labelled as an ontology layer, from Sources at the base to Agents and apps at the top L0 · SOURCESL1 · PIPELINESL2 · OBJECT TYPESL3 · LINK TYPESL4 · ACTIONSL5 · AGENTS & APPS StereobateStylobateColumnsEntablaturePedimentAcroteria
LayerWhat it holdsIn a logistics business
L5ActApps and AI agents, using the same actions as peopleControl tower on dispatcher screens; an agent that flags shipments stuck for 6+ hours
L4Reinvent · ActTyped actions with rules, approvals and write-backReroute shipment, Re-promise order, Reorder stock
L3UnderstandRelationships with meaning and directionOrder is fulfilled by Shipment; Truck carries Shipment
L2UnderstandThe nouns people already useCustomer, Order, Shipment, Truck, Hub
L1ConnectVersioned transforms, quality checks, entity resolutionMerge three customer records into one Customer
L0ConnectRaw systems, files, devices and APIs, untouchedERP and TMS databases, GPS feeds, Excel hub lists

Themes

Failed structures, rebuilt by a shared model

Each theme tells the same story through a different ruin: something great fell apart because its parts stopped connecting. All four run live on the stage above.

Live on the stage

Greek · 5th century BCE

Kintsugi Parthenon

A temple that fell, rebuilt the Japanese way, with gold in the cracks. The repairs stay visible on purpose. That is lineage: you can always see where a number was mended.

Columns → object types
Entablature → link types
Gold seams → provenance

Greek · recovered 1901

Antikythera Engine

The ancient Greek computer, pulled from a shipwreck as 82 corroded fragments. Rebuilt, it predicts eclipses. Every gear is a pipeline and every mesh between gears is a join.

Drive wheel → ingestion
Gear trains → transforms
Dial pointers → agents

Hiyaku · our mark

Hiyaku Lotus

The lotus grows clean out of muddy water, the way trustworthy knowledge grows out of messy sources. The infinity loop beneath it is the feedback cycle between actions and data.

Water → sources
Infinity loop → pipelines
Petal rings → layers L2 to L4

Myth · Genesis 11

Babel, Reversed

The tower failed when its builders stopped sharing a language. A shared ontology is that language. Every department keeps its own dialect; the layer above translates.

Tiers → layers
Scattered bricks → siloed systems
Beacon → one shared answer

Any industry

Same four primitives in every industry

Objects, links, actions and agents. Change the nouns and verbs and the core does not change. That is what makes it work in any industry: the business lives in the ontology, not in custom code.

IndustryObjectsLinksActionsAn agent that…
LogisticsShipmentTruckHubcarried byrouted viaRerouteRe-promisereroutes freight before a delay reaches the customer
RetailStoreSKUShopperstocked atbuysReorderMark downmoves stock before a store runs out
ManufacturingMachineBatchSupplier lotproduced onmade fromHold batchSchedule maintenancetraces a defect back to one supplier lot
Financial servicesAccountPaymentInvoicepayssettlesMatchHold paymentreconciles unmatched payments before month end
Energy & utilitiesAssetSiteCrewlocated atmaintained byDispatchIsolatesends the nearest crew before an outage spreads
AviationAircraftFlightPassengeroperatesbooked onSwap aircraftRebookrebooks connections when an aircraft is grounded

Build order

Built foundation first

The order matters. Each phase is useful on its own and is the footing for the next, the same way the temple goes up.

  1. Understand

    Ontology core and open spec

    Object, link and action types written as reviewable files in git. They compile to Iceberg tables, a GraphQL and REST API, and RDF for anyone who needs it.

  2. Connect

    Connectors and identity

    Connectors for common databases, spreadsheets and APIs. Entity resolution with blocking, match scores, and a human review queue for uncertain pairs.

  3. Understand

    Policy-carrying lineage

    Column-level lineage. Purpose and sensitivity markings propagate through every transform, so derived data inherits the strictest rule automatically.

  4. Act

    Actions and write-back

    Typed actions with validation, approvals, sagas and idempotency keys. Decisions go back to source systems with a full audit trail.

  5. Act

    Agents as colleagues

    Agents get roles and call actions through a tool interface. Same permission checks as staff, evaluation suites before rollout, every step replayable.

  6. Connect

    Federation and edge

    Sovereign nodes answer pushed-down queries without moving raw data. Offline-first edge nodes sync when a remote site's connection returns.

Contact us

Get started.

Tell us about one decision your team makes every day: a reroute, a stock move, a payment hold. We will show you how far toward autonomy it can go.

We use these details only to reply to you. See our privacy policy.