Skip to main content
Graph-native language

flowg.
Code is a typed dataflow graph.

A graph-native language. Programs are typed dataflow graphs that materialize to any language, lower to many silicons, and meter in picojoules. Port and edit without re-debugging. Spend less time, energy, and money.

→ any language
→ many silicons
pJ metered
load matmul norm out LOWER TO CPU 41 pJ GPU 9 pJ NPU 2 pJ ✓

One graph. Lowered to the cheapest silicon.

The language

One graph, many silicons

Write the program once as a graph, then lower it to the language and the chip that run it for the fewest joules.

A typed dataflow graph

Programs are graphs, not files. Types flow along the edges, so the structure is the program and the program is portable.

Materialize to any language

Lower the same graph to the language you need. Port and edit without re-debugging the logic each time.

Lowered to the cheapest silicon

One graph maps to many targets, CPU to accelerator, each operation metered in picojoules and routed to where it costs least.

The discipline

AI and compute, on a SWaP-2C budget

A program that knows its own shape can be placed for efficiency. flowg lowers the same graph to a SWaP-2C budget on whatever silicon is in front of it.

Size

One graph instead of a rewrite per platform.

Weight

Port without re-debugging. The logic travels, the boilerplate does not.

Power

Every operation metered in picojoules, edge by edge.

Cost

Lower to the cheapest silicon for the job, automatically.

Cooling

Work lands where it runs coolest, not just where it was written.

Program the graph

flowg-lang.dev →