Encode. Spike. Wire. Stimulate. Deploy.
Modular, hardware-agnostic libraries for spiking neural networks and bio-inspired computation.
Limen Neural is a solo research org building small, reusable building blocks for SNNs — encoding, dynamics, topology, training loops, GPU kernels, interchange (NIR), and runtimes. Work is experimental, agent-assisted, and released under permissive licenses for anyone who wants to try it.
This org is intentionally small. Broader experiments, tooling, quantization labs, viz, cloud notes, and one-off research live under my personal account @rmems so Limen-Neural stays a clean core surface for a solo maintainer.
I am a solo maintainer.
Starting September 2026, GitHub velocity here will slow down while I attend Western Governors University (WGU) for AI Engineering, and while I invest more time learning and manually reviewing code in:
- Rust
- Python
- Julia
- C# / .NET
Issues and PRs may sit longer. Review will be slower and more deliberate — that is intentional. The goal is deeper language fluency and better manual PR review, not maximum merge rate.
Thanks for patience if you are forking or building on these crates.
Limen Neural keeps hard repo boundaries. Each repository owns one concern.
| Layer | Purpose | Core repos |
|---|---|---|
| Encoding | Analog / feature streams → spike trains | axon-encoder |
| Neuron dynamics | Biologically grounded models + neuromodulation | neuromod |
| Topology & wiring | Graphs, delays, sparse synaptic maps | synaptic-mesh |
| Interchange | Framework-agnostic SNN IR (NIR) | nir-rs |
| Training loops | Offline / closed-loop plasticity | plasticity-lab |
| GPU kernels | Blackwell-oriented CUDA / Rust compute | myelin-accelerator |
| Runtime | Headless SNN inference daemon | brainstem-daemon |
Shared org CI and the profile README live in .github.
| Repository | Role |
|---|---|
| neuromod | Core SNN library: LIF, GIF, Izhikevich, FitzHugh–Nagumo, Hodgkin–Huxley, neuromodulators, plasticity hooks |
| axon-encoder | Sensory / feature → spike encoding (rate, delta, latency, population, …) |
| synaptic-mesh | Topology generators, axonal delays, sparse wiring |
| nir-rs | Pure-Rust NIR graph model + optional HDF5 .nir I/O |
| plasticity-lab | Domain-agnostic reward-modulated training loops around neuromod |
| myelin-accelerator | Low-level CUDA/Rust kernels (Blackwell / sm_120, ternary paths, routing/SAT experiments) |
| brainstem-daemon | Headless SNN inference runtime (soma-daemon) |
| Repository | Role |
|---|---|
| .github | Org profile + shared GitHub Actions workflows |
Personal and broader experimental repos (quantization bridges, MoE→SNN labs, telemetry, viz, HDL, trading research, agent tooling, cloud portfolio notes, Julia research apps, etc.) are not listed as Limen-Neural surface area. Browse @rmems if you are looking for those.
Examples of that wider sandbox include SAAQ / symbolic-regression work, hybrid MoE–SNN labs, gaming and mining telemetry collectors, spike visualization, SystemVerilog FPGA sketches, and multi-language playgrounds. They move faster, break more often, and are kept out of this org on purpose.
- Rust — performance-critical infrastructure: dynamics, encoding, wiring, IR, kernels, runtimes
- Julia — research-scale simulation and personal language practice (packages live under @rmems unless promoted here)
- Python / C# .NET — growing personal fluency for review, tooling, and broader ecosystem access (not a promise of first-class org packages yet)
- Hard boundaries — one repo, one concern; no cross-layer state duplication
- Portable & reproducible — CPU-first where possible; CUDA/FPGA paths explicit and optional
- Permissive licenses — MIT / Apache-2.0 unless a repo says otherwise
Python accessibility across open neuromorphic frameworks remains a longer-term interest as the core matures.
- Core SNN stack quality — stable APIs in
neuromod, encoders, mesh, and training loops - NIR in Rust — interchange without forcing a Python runtime
- GPU neuromorphic kernels — Blackwell-oriented paths in
myelin-accelerator - Hybrid SNN–LLM / quantization — mostly explored in personal
@rmemslabs, then folded back only when a piece is modular enough for this org
Shared workflows live in Limen-Neural/.github/workflows.
Typical gates (per language / repo):
- Rust — fmt, Clippy, build/test, docs, feature matrix, MSRV where configured
- Julia — package build, tests, formatting, coverage (shared templates; Julia packages are primarily under
@rmems) - Containers — optional reproducible images (GHCR / Docker Hub) where a repo opts in
Each library README shows its own badges. Agents and humans both benefit from failing early.
Limen-Neural/.github is the organization profile (what you are reading on https://github.com/Limen-Neural) and the home for shared GitHub configuration.
Libraries are permissively licensed. Use them, fork them, break them, rebuild them.
If you open issues or PRs:
- Prefer small, single-concern changes that respect repo boundaries
- Expect slow review from September 2026 onward (WGU + deliberate language practice)
- Read the target repo’s README for scope, commands, and ownership notes before proposing cross-crate redesigns
I am not running a large community project — just publishing modular experiments that I want to keep honest and reviewable by hand.
Solo experimental neuromorphic libraries — Limen Neural · maintained by @rmems · 2026
