Neural software runtime
Build, ship, and verify
neural reflexes.
Portable SNN modules for software, robots, and autonomous systems — local, deterministic, and replayable by design.
Compose neural behaviorsSimulate them liveExport them anywhereVerify exactly what they did
Compose neural behaviorsSimulate them liveExport them anywhereVerify exactly what they did
Compose neural behaviorsSimulate them liveExport them anywhereVerify exactly what they did
Compose neural behaviorsSimulate them liveExport them anywhereVerify exactly what they did
Autonomous systems are too dependent on oversized intelligence.
Many agents rely on large general-purpose models for tasks that need neither language nor deep reasoning.
Too muchcompute
Too muchlatency
Too muchinfrastructure
Too littledeterminism
Too littleauditability
It often needs a tiny,
specialized neural reflex.
A robot avoiding an obstacle, a game agent reacting to motion, or a software system detecting a critical event does not always need a giant model.
The pieces exist. The software layer does not.
Neural modules still have no shared way to be packaged, composed, exported, and verified.
Can't be Packaged
Versioned modules with typed interfaces — not notebook dumps.
Can't be Composed
Wire sensors → blocks → actuators into one portable stack.
Can't be Exported
Ship the same graph to .synapse, WASM, Rust, Python, or ROS2.
Can't be Verified
Replay the critical path with receipts — independent of the author.
SynapseVM turns neural circuits into portable software.
Hugging Face for SNNs — with composable neural control and verifiable execution.
Explore
Discover NeuroBlocks with provenance and interfaces.
Compose
Wire sensors → blocks → actuators in the Workbench.
Simulate
Run the exact Stack live — signal flow in one loop.
Compile · Run
Ship .synapse · WASM · Rust · Python · ROS2.
Verify
Replay the critical path with NeuroReceipts.
Neural modules with interfaces, not opaque model files.
Building blocks
NeuroBlocks & NeuroStacks
Typed neural functions you compose into versioned, exportable behaviors.
Deterministic100%replayable ticks
Workbench
Compose live
Typed node editor with validation before compile.
3D Simulator
Same binary
Test the modules you ship — not a mock.
NeuroReceipts
Verified path
Input → action, recorded
Package once — compose, simulate, export, and verify the same graph.
Robotics
Fast local reflexes for collision avoidance, visual motion, navigation, and stabilization.
Verification that follows the software.
Separate trust claims, instead of hiding them behind one green check.
Artifact verification
Confirm package, modules, versions, runtime, and graph.
The uploaded package is rebuilt and compared — identity, lockfile, and module bytes. A matching digest is comparison to expected bytes, not a publisher seal.
Runtime receipt
Record the exact execution context for a critical event.
A receipt binds the Stack to the tick: input, pre-state, outputs, and commands. It is evidence of what ran — not a certificate that the world matched.
Deterministic replay
Re-run the same Stack with the same input and state.
A fresh runtime replays the disclosed path. Match means the same graph, same input, and same state produced the same outputs — not that a robot moved.
External validation
Verify execution in an independent environment.
Another runtime — including the local Rust verifier for a signed module receipt — can witness the same claims. A missing independent result stays unverified; it is never a green check.
Public anchor
Commit compact roots for tamper-evident audit history.
Compact identity roots can be recorded so later auditors detect tampering. Inclusion is a separate claim — a local receipt is not a public record.
Artifact verification
Neural software should feel like software.
Connectomes, extracted circuits, trained SNNs, and synthetic models compile through the same runtime and verification layer.
synapse — zshlocal · deterministic
$
Start building
Give your software a nervous system.
Compose a NeuroStack locally. Simulate it live. Export it anywhere. Verify exactly what it did.