Under the hood

One environment. Point it at any company.

Vyom builds a single living causal model of how one specific company actually works, then runs a workforce of agents from it. The same engine covers construction, manufacturing, SaaS, logistics, healthcare and more. What changes per industry is configuration, never code. Everything here is built and verified on reality-grade synthetic companies with known ground truth. We are pre-seed, with zero live deployments, and we say that plainly, because known ground truth is exactly what makes these worked examples checkable.

The engine and the config

Why “any company” is real, not a slogan.

There is one engine. It ingests events, builds a graph of variables and the weighted causal edges between them, propagates the effect of a change through that graph, simulates an intervention before anyone commits to it, and runs agents that watch, reason and act. That engine does not know or care what industry it is looking at.

What a specific company brings is configuration, not new software: the variables that describe its business, the connectors into its systems, the external signals that move it, and the metrics each role watches. A construction GC and a SaaS company load different configuration and get the same reasoning. If a line of runtime code ever read if industry == “construction”, that would be a bug we would remove, not a plan we would ship. The vocabulary is per company. The math is shared.

Onboarding

A new company’s model, built in about thirty seconds.

The proof that this is one environment and not a set of vertical products is what happens when a company that fits no template shows up. You describe the business in plain English. Vyom builds its model on the spot: the variables that matter, the weighted causal edges between them, the levers someone can actually pull, the roles who own each metric, 10 to 16 entity types, and about 1,500 synthetic records to give it a running start. No template. No industry-specific code.

From that moment the same six modes run on it exactly as they run on Meridian Construction. There is no separate build step for a new industry, because there is no new industry code to write.

InputA plain-English description of the business
BuiltVariables, weighted causal edges, levers, roles
Seeded10–16 entity types · ~1,500 synthetic records
Time~30 seconds · no template, no industry code
The six modes

Six modes, one environment.

Everything below is an instance of the same six modes running on a different company. They do not change by industry. Meridian Construction’s rain-to-cash chain and a hospital’s no-show-to-revenue chain are the same six modes pointed at different variables.

Surfacing
Tells you what matters before you ask, and answers anything you do ask.
Simulation
Plays out the consequence of a decision before you take it.
Propagation
When something changes, it reaches everyone it affects at once, in their terms.
Execution
Does real work across native surfaces and existing systems, inside permissions and audit.
Memory
Keeps what the company learns so the knowledge doesn’t walk out the door.
Learning
Sharpens its own model of the business as it watches more of it happen.
The breadth

What is coverable today.

Eight industries are built as templates you can load now, each configuration for the same engine. Beyond the eight, any company can be onboarded from a plain-English description. On connectors, the honest status: construction and SaaS are wired (integrations built for read and, where it makes sense, write); the other six are modeled (the connector surface specified and reasoned over, with the live external hookup as the remaining step). The reasoning runs identically across all of them.

Construction · wiredB2B SaaS · wired Manufacturing · modeled3PL Logistics · modeledHealthcare · modeledHospitality · modeledReal estate · modeledRestaurant groups · modeled Your own · ~30s onboarding
Deep case studies

The same environment, followed end to end.

Four worked examples on named synthetic companies, wired-first. Open each for the full how, why and what. Because the true structure was authored, every claim is checkable against known ground truth.

What is real, what is outstanding

Said once, plainly.

Built and verified on synthetic companies with known ground truth

Everything above is built and verified against reality-grade synthetic companies whose true cause-and-effect structure we authored, so every claim traces back to a number we can check. Vyom is pre-seed and pre-revenue, with zero live customer deployments.

Two genuine outstanding items, everywhere: large-GPU pretraining, and live external hookups to real partners’ systems. Connector packs: construction and SaaS wired; the other six modeled. Full status →

Where to go next

Watch it run, or come build yours.

The interactive Explore page is the hands-on version: switch companies and fire a signal through the model yourself. Or start the conversation.

Request access