Vyana,
plainly.

There is a reason this is the first node and not a press release. We would rather start with a sentence we mean than a sentence that performs.

So: Vyana Compute is a research and engineering company. We are building the substrate that the next decade of intelligent systems will run on – from the model layer down, eventually, to the silicon. We are starting in New Delhi, with a small team. We expect to grow slowly and on purpose.

The rest of this note is the long version of those three sentences.

IThe name

Vyana is a Sanskrit word for one of the five vital currents that move through the body in the yogic and Ayurvedic traditions. It is the all-pervading one – the breath that does not localise. It is what is happening, quietly, between every other thing.

Compute is the precise opposite. Compute is the point at which something diffuse becomes legible. Where intent becomes work. Where current resolves.

We named the company for the seam between the two. The work we want to do lives there.

IIThe thesis

Here is what we believe, said in the smallest number of sentences we could find.

The next decade of intelligence will not be limited by ideas. It will be limited by the substrate underneath them.

Models converge. Methods diffuse. The frontier walks across labs in months. What stays scarce – what compounds – is the layer below: how compute is gathered, routed, observed, and resolved into useful work. That layer is treated as plumbing today. We do not believe it is plumbing. We believe it is the discipline.

IIIWhat we are building

Three concentric problems. One node.

Compute. Models that fit the metal they run on. Architectures, kernels, and runtimes that respect the physics of their substrate rather than abstracting away from it. There is more performance – and more reliability – to be found by reading the hardware closely than by stacking another framework on top of it.

Systems. Schedulers, fabrics, and orchestration that treat compute as a flow rather than a fleet. Most of the current shipped today is wasted in transit. We want to see the current end-to-end: where it enters, where it bends, where it dissipates, where it resolves.

Hardware. A long horizon. We expect to build silicon when the case is unambiguous and the rest of the stack is ready to receive it. Not before.

That order is deliberate. We earn the right to each layer by doing the one above it well.

IVWhat we are not

We are not a foundation-model company. We are not a cloud reseller. We are not a wrapper. We will not ship a chat interface.

We are a company about the substrate. We will publish the work. We will be specific about what we do and what we don’t.

If a problem can be solved with software someone else has already written, we will use that software. If it cannot, we will write it. If the problem is below software, we will go there too.

VThe horizon

We are starting in residence in New Delhi. A small team, then a few more. We have enough capital to spend the first eighteen months on research and the next eighteen on the first useful artifact. We are not in a hurry to be visible. We are in a hurry to be correct.

We will publish notes here as nodes – short, dated, indexed. Some will be technical. Some will be philosophical. Some will be very brief. Read them or don’t. They are the ledger.

VIA note on the mark

Two strokes of flow converging on a single node. The strokes are vyana – the all-pervading current. The node is the point of resolution.

Look closely. The strokes do not touch the node. There is a gap. That gap is intentional: motion is not allowed to collapse into structure, and structure is not allowed to consume motion. The aperture is where compute breathes.

We will keep that gap, in the mark and in the work.