Data basis: the GitHub Trending daily dump bot issue Rain1er/trending #430 (November 8, 2026, last updated 16:56 UTC) is the primary source, compared with the previous issue #429 to compute day-over-day gains. Cross-checked against kouweizhu/agents-radar #858 (01:45 UTC, previous-day snapshot) and happydog-intj #635. Repos with 40k+ total stars, long-running chart-toppers, slowing momentum and non-AI repos are excluded.
Excluded today: milind-soni/OpenMausBot (+1,986 to +1,437, down 28% day over day), sherlock-project/sherlock (100k+ and non-AI), vercel-labs/agent-browser (40k+), anthropics/skills (40k+, vendor directory), claude-quickstarts and mujoco (only about +20 each).
meta/astryx — +863 ⭐ today
https://github.com/meta/astryx · Total stars: 15,776 · Day over day: +185 to +863 (+366%) · TypeScript · MIT
Astryx is Meta’s open-source design system (in beta), built as an agent-ready UI kit that people and AI coding agents use the same way. It sits on React 19+ and StyleX and ships as @astryxdesign/core (components, themes, utilities), @astryxdesign/cli (docs, templates, scaffolding, codemods), @astryxdesign/build (StyleX source-build plugins) and @astryxdesign/theme-* packages. Pre-built CSS and typed components mean no build plugin is needed, and the repo includes AGENTS.md and CLAUDE.md so agents can query docs directly through the astryx docs CLI. Key features:
- 150+ accessible React components, dark mode and ready-made templates
- Seven themes (neutral, matcha, gothic, y2k and more) built as CSS custom property overrides
- className overrides with Tailwind, CSS modules or plain CSS, plus swizzle to eject a component’s full source
- Patterns for common workflows such as table pages and form wizards
Daily gains jumped more than fourfold in one day, reflecting demand for design systems that AI agents can read and use reliably in the vibe-coding era.
Practical takeaway: A solid, consistent UI foundation for teams building internal admin tools and dashboards with AI coding agents.
Tags: #DesignSystem #React #StyleX #AgentReadyUI #Accessibility
ranxianglei/billion-context — +460 ⭐ today
https://github.com/ranxianglei/billion-context · Total stars: 4,982 · Day over day: +185 to +460 (+149%) · TypeScript · LICENSE file (type not shown; the paper is MIT)
billion-context is a context-compression plugin for long-running AI coding sessions, pitched as month-long sessions on a 100K window with 5x fewer tokens. A proxy (bili) sits between the agent and the model API: it parses each request, compresses the conversation, injects context-management tools, forwards to the real API and rewrites the streaming response. Rather than hard truncation, the model decides when and what to compress, and compression is incremental, reversible and designed not to break prompt-cache prefixes (the engine lives in a separate acp-kernel repo). Key features:
- Injected tools: compress, decompress, search_context and acp_status
- Anthropic and OpenAI request formats, with native plugin, launcher (bili <client>) and URL-prefix integration modes
- Web UI on localhost:8787 (dashboard, sessions, logs, config) and prefix-cache hit-rate reporting
- Support for Claude Code, Codex, Gemini CLI, opencode, Aider, Copilot and more, plus a Docker image
Interest is rising as token costs and context limits become real bottlenecks for coding agents (benchmark figures are the project’s own claims).
Practical takeaway: Teams running long agent jobs can pilot it as a proxy to cut API spend and keep sessions continuous.
Tags: #ContextCompression #LLMProxy #CodingAgents #PromptCaching #TokenSavings
dsx-ai-factory/nvke — +345 ⭐ today
https://github.com/dsx-ai-factory/nvke · Total stars: 752 · Day over day: +125 to +345 (+176%) · Go · Apache-2.0
NVKE (NVIDIA Kubernetes Engine, NVIDIA copyright) is a preview management plane for running managed Kubernetes on your own NVIDIA GPU bare metal. A management cluster runs the NVKE plane and infrastructure controllers, then creates and operates customer workload clusters through the NVIDIA Infra Controller (NICo), Cluster API and a node agent. Each workload cluster gets its own control plane, etcd and node pools, making NVKE a self-hosting factory for Kubernetes clusters. Key features:
- Isolated GPU clusters on demand via the NVKE API or nvkectl CLI
- Lifecycle management: node-pool scaling and deletion, Kubernetes upgrades, node replacement
- TPM-based node attestation with SPIFFE/SPIRE and optional per-cluster etcd encryption
- A local evaluation path that needs no GPU hardware (Linux VM or macOS via Colima)
It reflects growing demand from GPU clouds and AI-factory operators to run hyperscaler-grade GPU Kubernetes themselves. API authentication is off by default in preview, so it is not recommended for production yet.
Practical takeaway: A reference architecture for neoclouds and in-house AI infra teams that want to carve GPU clusters out for multiple teams or customers.
Tags: #Kubernetes #GPUInfrastructure #BareMetal #NVIDIA #AIFactory
Leave a Reply