GitHub Trending — November 8, 2026: agents, infrastructure, and context

The trending repositories this week reveal a clear pattern: the AI ecosystem is shifting from isolated models to complete agent systems that must navigate, execute, and manage real computational resources. The nine most popular repositories fall into three pillars that define the infrastructure for the next generation of AI tools.

Agent browsing and control

agent-browser (Vercel Labs, Rust) — 812 stars today, 47,470 total https://github.com/vercel-labs/agent-browser

agent-browser is a native Rust CLI that gives an AI agent a browser it can drive one command at a time. Instead of relying on generic automation libraries like Selenium or Playwright, it was built from scratch with a client-daemon architecture that uses the Chrome DevTools Protocol directly. The `snapshot` command returns the page as compact text with refs (`@e1`, `@e2`) for each element — saving 200–400 tokens of context per call versus 3,000–5,000 in JSON. The browser persists between commands, so each step is just a short shell call.

What sets agent-browser apart is its focus on agents, not human automation. It ships version-matching skills, works with Claude Code, Cursor, GitHub Copilot, OpenAI Codex and others, and supports persistent sessions with auth state, cookies, and storage — essential for agent workflows that require multiple steps. The Vercel team also integrated cloud provider support like Browserbase and Browserless, making it flexible for production environments.

OpenMausBot (Milind Soni, TypeScript) — 1,437 stars today, 18,073 total https://github.com/milind-soni/OpenMausBot

Inspired by xAI's Grok Bot, OpenMausBot recreates the idea of a chat app where every conversation is a real agent — Claude, Codex, or Grok running locally — with its own personality, model, cloud computer, and connected apps. The crucial difference: it's open-source, runs on your own machine, uses your own logins, and doesn't require expensive subscription plans.

Each bot runs through locally installed CLIs (`claude`, `codex`, `grok`), with a harness server at `127.0.0.1:8799` that manages processes and events via SSE. Transcripts, keys, and events stay in `~/.openmausbot`, not the cloud. The project supports 500+ apps via Composio, cloud computers via Boat, native local machine control via Cua Driver, and hosted voices via ElevenLabs or Fish Audio.

The project gained traction because it addresses a real need: users who want the multi-agent experience of Grok Bot but don't want to be locked into xAI's closed ecosystem.

Agent infrastructure

NVIDIA Kubernetes Engine (NVKE) (DSX AI Factory, Go) — 345 stars today, 787 total https://github.com/dsx-ai-factory/nvke

NVKE is a management layer that allows operators of NVIDIA GPU data centers to offer Kubernetes as a managed service to their customers. Each workload cluster gets its own control plane, etcd, and node pools. The operator manages the management plane; customers access clusters via the Kubernetes API and `kubectl`.

The system provisions bare-metal servers through NICo (NVIDIA Infra Controller), enabling physical infrastructure management via APIs rather than manual runbooks. The GPU Operator installs drivers, Container Toolkit, and device plugins automatically. The project is part of NVIDIA's DSX OS stack and licensed under Apache 2.0.

The analogy is AWS EKS but in open source and GPU-focused: instead of an operator assembling a management system from open-source and commercial components, NVKE delivers it as a single product. It's in preview and NVIDIA doesn't yet recommend it for production. According to the team, the system doesn't self-manage — it doesn't update, scale, or replace its own nodes automatically.

skills (Anthropic, Python) — 793 stars today, 189,429 total https://github.com/anthropics/skills

Anthropic's official implementation of the Agent Skills standard for Claude. Skills are folders of instructions, scripts, and resources that Claude loads dynamically to perform specialized tasks — from creating documents with brand guidelines to analyzing data using specific organizational workflows.

The repository includes Anthropic's internal skills powering production DOCX, PDF, PPTX, and XLSX generation. The spec is in the repo, aiming to create a shareable capability ecosystem for AI agents. The repository is a reference for developers who want to create custom skills and understand the patterns of how agents can be extended in a standardized way.

claude-quickstarts (Anthropic, Python) — 28 stars today, 19,337 total https://github.com/anthropics/claude-quickstarts

Reference project collection for the Claude API, covering customer support agents through autonomous coding agents. Recent quickstarts include Computer Use, Browser Use, and Managed Agents with patterns like token-by-token streaming, vault credentials, and event deltas.

The importance lies in being a starting point for teams building on the Claude API. The Managed Agents examples are particularly relevant — demonstrating scheduled agents that read Slack and GitHub, or production error fixing triggered by webhooks.

Infrastructure tools and simulators

Sherlock (Sherlock Project, Python) — 2,372 stars today, 103,022 total https://github.com/sherlock-project/sherlock

Sherlock is an OSINT tool that searches for usernames across 400+ social networks. With 2,934 commits and an active contributor community, it's a reference tool for security researchers and threat investigators who need to map a target's digital footprint across multiple platforms simultaneously.

Astryx (Meta, TypeScript) — 863 stars today, 16,250 total https://github.com/meta/astryx

Open-source design system that grew inside Meta over eight years, becoming the company's most-used and largest design system — powering 13,000+ apps. With 150+ accessible components, brand-level theming, dark mode, and a CLI, Astryx is built on React 19+ and StyleX. The differentiator: both humans and AI assistants build with the same tooling.

MuJoCo (Google DeepMind, C++) — 26 stars today, 16,319 total https://github.com/google-deepmind/mujoco

MuJoCo (Multi-Joint dynamics with Contact) is a general-purpose physics engine used in robotics, biomechanics, machine learning, and simulation. The repo includes bindings for Python, JavaScript/WebAssembly, and Unity. It's the backbone for countless reinforcement learning environments and has been extensively used in robotic locomotion research.

billion-context (ranxianglei, TypeScript) — 460 stars today, 5,048 total https://github.com/ranxianglei/billion-context

billion-context is a proxy that sits between any AI agent and its model API, rewriting Anthropic/OpenAI streams with incremental compression. The model decides when and what to compress into high-fidelity summaries — not hard truncation. In production, the project sustained sessions of thousands of calls (up to 12,000+) on a 204,800-token window, averaging ~126K tokens per call and peaks of ~1.52B cumulative tokens.

The system injects four context-management tools: `compress` (fold a range into a summary), `decompress` (restore a compressed range), `search_context` (search compressed summaries), and `acp_status` (usage overview). Compression costs ≤2%, and healthy sessions maintain a 95–97% prefix-cache hit rate.

Trend: the ecosystem consolidates

The nine repositories on this ranking draw a coherent picture of what the developer community is building. The dividing line between "AI models" and "AI infrastructure" is disappearing — it's no longer enough to have a good model; you need it to navigate the web, manage GPU clusters, compress its own context, and operate in a secure environment.

agent-browser and billion-context represent the practical answer to the context problem: while the first optimizes output to save tokens, the second compresses input to extend sessions. Together, they cover both ends of the agent pipeline.

NVKE and skills reflect verticalization: NVIDIA is becoming an infrastructure platform (not just hardware), while Anthropic standardizes how third-party agents are extended. It's a movement toward infrastructure-as-code, not just infrastructure-as-API.

The question is: when the next GitHub Trending comes out, how many of these projects will still be at the top? The agent ecosystem is moving so fast that the trending list is already a historical record — not a guide to where things are headed.

Sources: GitHub Trending, agent-browser changelog, heise online

✓ Independent sources cross-checked and verified before publishing