9 November 2026

Find social media accounts from a username across 400+ networks

Sherlock searches one or more usernames across more than 400 social networks from the command line. Found accounts are saved to a text file per username.

• It installs with pipx, pip, uv, or Docker.
• Results can also be exported to CSV or XLSX, and a search can be limited to chosen sites or sent through a proxy.

Managed Kubernetes service on your own NVIDIA GPU infrastructure

NVKE is a management plane for creating and operating Kubernetes clusters on your own NVIDIA GPU infrastructure. It is a preview and not recommended for production.

• Each workload cluster has its own control plane, etcd, and node pools, and is requested through the NVKE API or nvkectl.
• Nodes prove their identity through SPIFFE/SPIRE with a TPM before they receive join credentials.
• API authentication is off after installation and must be configured by the operator.

A team of AI bots in a chat app, running locally on agents you already have

OpenMausBot is an open-source project inspired by Grok Bot: a roster of bots you chat with, each with its own model, computer, and connected apps. It is licensed under Apache 2.0.

• Bots run on the claude, codex, and grok CLIs installed on your machine, using your existing logins.
• One small harness server on 127.0.0.1 owns every agent process, and transcripts are stored in ~/.openmausbot.
• Bots can use a cloud Linux desktop, a Local VM, or your own computer after opt-in, plus 500+ apps through Composio.

Starter projects for building applications with the Claude API

Claude Quickstarts is a collection of example projects you can build on for applications using the Claude API. It needs an API key from console.anthropic.com.

• Examples include a Customer Support Agent, Financial Data Analyst, Computer Use Demo, and Browser Use Demo.
• It also has an Autonomous Coding Agent built on the Claude Agent SDK with a two-agent pattern.

A Rust CLI that gives AI agents a browser

agent-browser gives an AI agent a browser it drives one shell command at a time. Pages are read as compact text with a ref such as @e1 for each element.

• The agent calls snapshot, acts on a ref, then reads the page again to check the result.
• The browser stays open between commands, so each step is one short shell call.

Meta's open-source design system for people and agents

Astryx is an open-source design system with 150+ themeable React components that works without a build plugin. It is currently in Beta.

• A theme is a set of CSS custom property overrides, so there is no need to fork or wrap components.
• It is built on React 19+ and StyleX, and styles can be overridden with Tailwind, CSS modules, or plain CSS.
• swizzle ejects a component's full source into your project.

Anthropic's public repository of Agent Skills for Claude

Skills are folders of instructions, scripts, and resources that Claude loads dynamically for specialized tasks. This repository holds example skills, the specification, and a template.

• It includes the docx, pdf, pptx, and xlsx document skills used in Claude's document features.
• It can be added as a Claude Code plugin marketplace and installed one skill set at a time.
• The skills are for demonstration and education, and should be tested before use.

Physics simulator for robotics and research, from Google DeepMind

MuJoCo is a physics engine for articulated structures interacting with their environment, with a C API. It is maintained by Google DeepMind.

• Data structures are preallocated by the built-in XML compiler, and the runtime simulation module is tuned for performance.
• It has Python bindings and a Unity plug-in, plus a WebAssembly version at live.mujoco.org.
• Tutorial notebooks run on Google Colab, including LQR and MJX.

Context-compression plugin for very long agent sessions

billion-context sits between an agent and its model API and rewrites Anthropic/OpenAI streams with acp-kernel compression. The model itself decides when and what to compress.

• The proxy injects four context-management tools, and compress runs server-side.
• Summaries are written in small ranges, can be restored on demand, and the cache prefix stays intact.
• The README reports a 95 to 97 percent prefix-cache hit rate in a healthy session.