🧠 Spikeling

A neuromorphic domain-specific language and runtime for spiking neural networks, by Gavin Branaa. You describe a network of spiking neurons and synapses in a compact .spk file, and run that one description on any of four backends:

• an interactive Python runtime (development + STDP learning),
• generated C (embedded / production),
• generated Verilog (FPGA / hardware simulation), and
• GDScript (give a Godot game a live spiking "mind").

One language, four targets.

Quick start

cd core
python -m core                     # run the default example, interactively
python -m core path/to/network.spk # run your own .spk network

Full DSL reference, runtime details, and the C-compilation path are in core/README.md.

The .spk language in 30 seconds

neuron LeftMic  threshold=110 leak=5 type=LIF
neuron RightMic threshold=110 leak=5 type=LIF
neuron Motor    threshold=80  leak=3 type=LIF

connect LeftMic  -> Motor weight=0.8
connect RightMic -> Motor weight=0.8

action Motor -> [MOTOR_FIRE]

refractory=400ms
learn=STDP rate=0.01

(The GDScript backend uses a compact variant — synapse SRC -> DST weight=N — documented in godot-runtime/.)

Neuron types

Repository map

Not included here: a separate, unrelated project is kept out of this repository via .gitignore.

Agent orchestration as an SNN

An experimental line of work tests whether a spiking neural network can serve as the control layer for a multi-agent pipeline — routing tasks, managing concurrency, and arbitrating conflicts through spike-based inhibition rather than classical scheduling logic.

The honest summary: the SNN routing and inhibition primitives work, but the scheduling and arbitration results did not beat well-known classical baselines.

License

MIT (source code). See LICENSE.