Today's GitHub trending charts tell a story of decentralization: developers are pushing AI further down the stack — onto mobile devices, IoT sensors, and local machines — while simultaneously reaching for more composable, modular application architectures. The cloud remains the baseline, but the interesting bets are happening at the edge.
cactus-compute/needle: A 14MB Foundation Model Built for the Edge
Needle is a compact foundation model designed to run on devices where memory and compute are at a premium — mobile hardware, IoT endpoints, and embedded systems. At just 14 megabytes, it challenges the assumption that capable AI requires heavyweight infrastructure. With 443 stars today, it's drawing attention from developers who need local inference without cloud round-trip latency or the privacy concerns that come with sending data to a remote endpoint.
github.com/cactus-compute/needle
unslothai/unsloth: Fast Local LLM Training Without the Hardware Tax
Unsloth is a Python toolkit that makes training and running large language models and diffusion models significantly faster and cheaper on local hardware. It supports a range of cutting-edge models and is tuned for memory efficiency, making fine-tuning accessible to developers who don't have access to high-end GPU clusters. 572 stars today suggest it's filling a real gap between expensive cloud training and practically impossible local training for most developers.
cordiverse/cordis: Composability as a First-Class Architecture Primitive
Cordis describes itself as a "meta-framework of spatiotemporal composability" — a TypeScript framework where components are designed to be composed, extended, and swapped at any layer of the stack. 720 stars in a single day is a strong signal that TypeScript developers are hungry for something that treats modularity as a fundamental guarantee rather than a convention. Think less "plug-in system" and more "every piece was always meant to be replaced."
basecamp/omarchy: Basecamp's Opinionated Take on Linux
Omarchy is Basecamp's curated Linux environment — opinionated defaults, a coherent aesthetic, and a batteries-included setup for developers who want to skip the distro assembly process. 270 stars today is modest but the Basecamp pedigree will draw attention from developers who trust their design sensibility. It's less about what Omarchy does and more about whether you trust the opinions of the team behind Rails, HEY, and Basecamp itself.
ToolJet/ToolJet: Self-Hosted Internal Tools with AI Built In
ToolJet is an open-source platform for building internal dashboards and business applications, and its 452-star day reflects continued momentum as teams look to self-host rather than depend on vendors like Retool. Recent releases added AI capabilities for connecting data sources and wiring up logic without writing full-stack applications from scratch. For engineering teams spending too much time on admin tooling, it's a compelling escape hatch from both vendor lock-in and the maintenance burden of custom builds.