Documentation

Platform overview

SynapseVM is software infrastructure for building, running, and proving tiny deterministic neural controllers — the reflex layer machines need when a large model is too slow, too costly, or too opaque.

What the platform does

SynapseVM turns neural modules into portable software packages. You select versioned NeuroBlocks, compose them into a NeuroStack, run that stack in a deterministic runtime (SynapseVM), and — for critical decisions — emit a NeuroReceipt that another party can replay bit-for-bit.

NeuroLab is the workbench on top of that stack: Discover to find modules, Compose to wire them, Simulate to watch them live, and Verify to check evidence.

┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  Discover   │───▶│   Compose   │───▶│  Simulate   │───▶│   Verify    │
│  find       │    │  wire &     │    │  run live   │    │  replay     │
│  modules    │    │  package    │    │  loop       │    │  evidence   │
└─────────────┘    └──────┬──────┘    └─────────────┘    └─────────────┘
                          │
                          ▼
                   .synapse package
                   (NeuroStack artifact)

Why it exists

Planners and large models decide goals. Controllers must still react in milliseconds: dodge, brake, stabilize, veto an unsafe command. Those reactions need to be local, cheap, and auditable.

SynapseVM targets that layer. It does not replace your planner. It packages the neural reflexes that must keep working when the network is gone and that must leave a trail when something critical fires.

Design constraints

Determinism — same stack, same input, same state ⇒ same outputs across conforming runtimes.
Local-first — control ticks do not wait on a chain, oracle, or payment rail.
Honest claims — a receipt proves a disclosed computation path, not physical actuation or publisher virtue by itself.
Typed composition — illegal port wiring fails the graph checker before compile.