Today's GitHub trending is less about building new models and more about making the existing ecosystem accessible: running models locally that actually fit your GPU, routing across hundreds of providers without vendor lock-in, and surfacing real satellite data through a browser globe. The infrastructure layer of AI development is getting more interesting than the headline models.
bilawalsidhu/gods-eye-view: Real Satellite Data on a Photorealistic Globe
This JavaScript project turns a browser tab into a live open-source spatial intelligence viewer — real satellite positions, orbital paths, and ground tracks rendered on a photorealistic 3D globe. The "spy satellite simulator" framing is attention-grabbing, but what's actually interesting is how much verified public data is now accessible and renderable in a browser without a backend. With 25,600+ stars and nearly 1,800 gained today, it's resonating well beyond satellite enthusiasts — it's a reference point for what client-side data visualization can pull off in 2026.
github.com/bilawalsidhu/gods-eye-view
AlexsJones/llmfit: Find Which Models Actually Run on Your Hardware
A Rust CLI that interrogates your hardware and tells you which models — across hundreds of providers and local sources — will actually fit and how well they'll perform. The problem it solves is real: discovering that a model won't fit in VRAM after a 20-minute download is a workflow killer. llmfit moves that check to before the download, outputting a ranked compatibility list in one command. At 36,000 stars with growing daily momentum, it fills a gap that model cards and provider dashboards have never addressed cleanly.
diegosouzapw/OmniRoute: One Endpoint for 1,200+ Models
OmniRoute is a free, MIT-licensed AI gateway that routes requests across 352 providers — 150+ of which are free — through a single API endpoint. The pitch is provider-agnostic development: write your code once against the OmniRoute interface and swap providers without touching your integration. At 64,500 stars and a TypeScript codebase, it's already a significant community project. The 1,200+ model count and "150+ free" tier framing suggest it's particularly aimed at developers who want to experiment across the landscape without juggling multiple API keys and SDKs.
github.com/diegosouzapw/OmniRoute
alsk1992/CloddsBot: Autonomous Trading Across 1,000+ Markets
An open-source AI trading agent written in TypeScript that operates across more than 1,000 markets with fully self-hosted capabilities. The architecture puts the agent entirely under the operator's control — no cloud dependency for execution — which addresses one of the bigger trust problems with autonomous finance tools. With nearly 1,900 stars and 277 added today, it's early-stage but climbing fast. Projects like this are useful less as production trading systems and more as public blueprints for autonomous agent architecture in high-stakes, latency-sensitive domains.
freestylefly/awesome-gpt-image-2: 530+ Industrial Image Prompts
A prompt library with over 530 curated examples and industrial templates aimed specifically at GPT image generation. Unlike general-purpose prompt guides, this one organizes examples by professional use case — product photography, architectural rendering, marketing materials — making it easier to find a production-ready starting point for commercial work. At 31,200 stars, it reflects the continued demand for well-structured prompt references as image generation moves from experimental to production creative workflows.