GitHub Trending is a window into what developers are building and exploring in real time. This week, the most popular repositories reflect clear trends: AI models on constrained devices, specification-driven development, and visual diagramming tools for Claude Code.
1. cordiverse/cordis — TypeScript
⭐ 616 today / 3,877 totalCordis is a meta-framework for spatiotemporal composability that enables applications that evolve dynamically over time. It solves the classic problem of applications that need to maintain contextual state across multiple interactions — essential for AI agents operating in long sessions.
2. cathrynlavery/diagram-design — HTML
⭐ 1,619 today / 18,260 totalDiagram-design offers 29 editorial diagram types for Claude Code. Self-contained HTML files with SVG, no shadows, no Mermaid aesthetics. The project addresses a real need: professional diagrams that don't look generically AI-generated.
3. cursor/plugins — TypeScript
⭐ 152 today / 2,898 totalCursor's plugin specification and official plugins. As Cursor consolidates as the dominant AI editor, the plugin ecosystem becomes as important as VS Code extensions became for the JavaScript ecosystem.
4. cactus-compute/needle — Python
⭐ 551 today / 5,926 totalNeedle is a 14MB foundation model for constrained devices: phones, wearables, smart home, and robots. The tiny footprint enables offline execution on hardware that couldn't run AI before. It's the Model T equivalent for the LLM world — accessible where others fail.
5. unslothai/unsloth — Python
⭐ 435 today / 71,836 totalLocal UI to run and train LLMs and diffusion models, including Qwen3.8, Kimi K3, MiniMax-H3, Gemma 4, DeepSeek-V4, FLUX and more. With 71k stars, it's GitHub's most popular local training repository.
6. public-apis/public-apis — Python
⭐ 2,476 today / 459,674 totalA collective list of free APIs. With almost 460k stars, it's GitHub's most starred repository that serves as a reference for developers looking for APIs to integrate into their projects.
7. MakazhanAlpamys/Soup — Python
⭐ 303 today / 1,515 totalSoup allows training LLMs from a single YAML file. Layer streaming trains an 8B model on a 4GB laptop GPU. The philosophy is simple: democratize fine-tuning by eliminating operational complexity.
8. github/spec-kit — Python
⭐ 901 today / 128,983 totalToolkit for Spec-Driven Development. With nearly 129k stars, GitHub's Spec-Kit represents a clear trend: the shift from "writing code" to "specifying intentions" and letting AI generate the implementation.
9. megadose/holehe — Python
⭐ 389 today / 13,023 totalHolehe checks if an email is used on different sites like Twitter, Instagram, and retrieves information through the forgotten password function. An essential OSINT tool for security professionals.
10. altic-dev/FluidVoice — Swift
⭐ 311 today / 10,217 totalFluidVoice is the fastest macOS Dictation app with on-device STT and custom trained AI enhancement. A local Wispr Flow alternative. Windows and iOS waitlist open.
This week's trend
This week's repositories reveal three clear trends:
1. AI on constrained devices: Needle (14MB) and FluidVoice (local STT) show that AI is descending from datacenters to devices. The future is running intelligent models on hardware that consumes watts, not kilowatts.
2. Specification-driven development: Cursor Plugins and Spec-Kit indicate that developers are migrating from writing code to specifying intentions. AI generates the implementation, humans define what.
3. Tools for Claude Code: diagram-design with 1,619 stars shows that the Claude Code tool ecosystem is maturing. If Claude becomes the dominant model for development, tools around it will be as valuable as the model itself.
What to watch: the race to run AI on increasingly constrained devices will continue. Who can train useful models on 4GB GPUs will truly democratize AI access — not just access, but local privacy and connectivity independence.
Sources: GitHub Trending, public-apis/public-apis, github/spec-kit
✓ Independent sources cross-checked and verified before publishing
