Older GPUs welcome
Start with the evidence-backed RX580/RADV path. Add Vulkan, CUDA, ROCm, Metal, or another backend through explicit capability gates.
GPU contribution guide ↗OPEN HARDWARE · OPEN QUESTIONS
Local AI,
kept in orbit.
A transparent, community-built model lab and runtime for bringing capable local AI to older systems.
CPU reference first · Vulkan acceleration second · evidence always visible
EXPERIMENTALBuilding in public
RX580 / RADVValidated target
CPU → VULKANClear fallback contracts
COMMUNITYBring your hardware
THE MISSION
Useful hardware has
more to give.
Models stay inspectable
New models, tokenizers, adapters, and quantization paths come with provenance, shape contracts, fixtures, and honest limitations.
Model contribution guide ↗Experiments stay reproducible
Commands, hardware, datasets, checksums, metrics, failures, and status are part of the result—not an afterthought.
Experiment protocol ↗A SHARED FLIGHT PLAN
Build it together.
Show the working.
Spaceslug is for people with curious machines, constrained budgets, and useful ideas. Whether you are profiling a shader, adapting a model, documenting a failure, or taking your first local-AI step, your contribution can move the mission forward.
Read the contribution guide ↗