Releases

Syncular v2 Rust client core on rusqlite, implemented from SPEC.md alone (language-neutrality POC, stage 2)

The generic layer resolver over the forensic-vfs contracts: sniff a byte source, match a registered prober, and descend container/volume/filesystem layers until a filesystem mounts — the evolving detection behavior, firewalled from the frozen forensic-vfs contract leaf

Read-only forensic virtual-filesystem contracts: the ImageSource positioned-read byte source, layered PathSpec locators, and the FileSystem navigation trait — the KNOWLEDGE leaf every disk/container/filesystem reader in the fleet implements

On-device runner for exported learned policies (RL/VLA MLPs) — inference only, WASM-clean. Interop tier: run trained weights, don't rewrite training.

Scientific and sci-fi color palettes from ggsci for Rust

Core deterministic NTP protocol engine for OpenNTPD-rs (no I/O, no host clock)

Syncular v2 SSP2 wire codec, implemented from SPEC.md alone (language-neutrality POC)

Scheme evaluator for the ma actor network — parser, evaluator, and SchemeCtx host trait

2D incompressible Navier-Stokes (MAC projection) for fluid–robot interaction. WASM-clean.

Differentiable Discrete Elastic Rods for cables, tendons, and continuum/soft robots. WASM-clean.

Robot kinematic optimization in Rust — FK, analytic Jacobians, Levenberg-Marquardt IK. WASM-clean.

Differentiable optical-tactile (GelSight/DIGIT) sensor simulation — gel deformation → photometric image. WASM-clean.

Differentiable FEM thin-shell cloth (StVK membrane + bending) for physical AI. WASM-clean.

Differentiable 2D Material Point Method (MLS-MPM) for soft/elastic/granular material. WASM-clean.

YAML/file-backed DotRegistry for ma-zscheme

IPFS-backed DotRegistry for ma-zscheme

Jerk-limited, time-optimal trajectory generation in Rust — a port of Ruckig's MIT Community S-curve. WASM-clean, no deps.

Rust-owned NNRP conformance fixtures, golden vectors, and adapter helpers.

Runtime: frame scheduler, event system, subsystem manager; ties every other crate into the main loop.

C-compatible NNRP FFI surface with value handles, event polling, and native library packaging.

Rigid body dynamics built on the GAC: broad/narrow phase collision, constraint solving and integration.

High-level rendering: render graph, scene extraction, visibility/culling, lighting, materials, camera and post-processing.

Low-level GPU abstraction (device, command queues, buffers, textures, shaders, pipelines, synchronization). Knows nothing about scenes or materials.

WASM primitives for NNRP framing, structured probe metrics, and deterministic transport selection.

Batch execution kernels for GAC transforms (Motor3) on CPU-SIMD / GPU compute. The adapter between the pure math in gac-core and the dispatch runtime in compute-runtime — neither depends on the other.

Low-level compute dispatch abstraction (CPU SIMD backends, GPU compute queues/buffers). Knows nothing about physics, animation or rendering.

Native WebSocket transport provider for NNRP binary frames.

Compute dispatch runtime: device/context, buffers, and the ComputeKernel interface domain crates (gac-compute, future particle-compute/physics-compute/...) implement against. Owns no algorithms of its own.

Shared low-level wgpu device/buffer/shader/compute-pipeline mechanics underneath graphics-driver and compute-driver. Knows nothing about scenes, materials, or compute-dispatch scheduling policy.

TCP transport provider package for NNRP framed runtime transport and listener slots.