Releases

Shared Mob host integration with MCP surface adapters

Compatibility re-export for Meerkat Mob adaptive internals

Mobpack format types and validation primitives

Runtime for slave apps (a Laravel-inspired Rust framework for MCP-first apps): server lifecycle, config, logging, Turso database and migrations, password hashing and bearer tokens

MoosicBox schema package

Rust SDK for programmatic control of the GitHub Copilot CLI via JSON-RPC.

Functional Mock-up Interface (FMI) for Rust

MoosicBox embedded historical SQL migrations

Policy-driven six-tier cache orchestration with an explicit HPA data-continuity contract

RLX — small ML compiler + runtime for transformer inference and training. JAX-shaped IR, autodiff, vmap, on top of CPU / Apple Silicon (Metal / MLX) / NVIDIA / AMD / TPU / wgpu backends. This is the prelude crate; pulls in rlx-ir / rlx-opt / rlx-runtime and re-exports the common types.

A desktop keeper for uv virtual environments: track, sync and open the environments you already have.

Flow-map generative policies with Flow Map Q-Guidance (FMQ) and QGBS for RLX.

Fetches and embeds a pinned uv binary at build time, for Bombadil.

PLAN L5 — uniform benchmark harness for RLX backends + patterns

Moment-encoded lattice Boltzmann (HOME-LBM) for RLX — D2Q9/D3Q27 reconstruction, collision, and periodic streaming as rlx graphs.

Force density method (FDM) for RLX — pin-jointed form-finding ported from jax_fdm.

A fast, native Wayland status bar for Hyprland

ONNX inference for RLX — native compile by default, optional ORT fallback

Offline weight bake for RLX — specialize params, fold weight-only compute, emit a single *.rlx artifact

HyperChad documentation site framework

HyperChad package

RLX text — tokenizer wrappers, chat templates, sampling. Public surface for downstream LM apps.

Parametric UMAP on RLX — full fit/transform API and k-NN building blocks.

HyperChad state management package

HyperChad App package

HyperChad Vanilla JS HTML renderer package

Service for Verifiable Trust Agents operating in Verifiable Trust Communities

Parameter-efficient adaptation on rlx graphs: LoRA, IA3, AdaLoRA, DoRA and OFT, matching HuggingFace PEFT

Safe, typed Rust wrappers for NVIDIA FlashInfer's inference-serving kernels: batched paged-KV attention decode, decode-time KV-cache append, cascade / prefix-cache attention-state merge, and sort-free top-K / top-P / min-P sampling. The canonical vLLM-style serving surface for the baracuda CUDA stack. Apache-2.0 (FlashInfer upstream).

Data-pattern recording harness — inject stat taps into op graphs, sweep synthetic distributions, and mine exploitable structure for specialized kernels