Packages

UI-TARS is an open-source multimodal agent for automated GUI interaction. It uses a vision-language model to parse/ground visual observations and generate structured action instructions that can be translated into automation code (e.g., PyAutoGUI) to operate desktop/mobile UIs.

A Claude Code “skill” that orchestrates multiple AI coding agents to execute parallel development tasks, using tmux panes, background processes, and/or MCP. It deploys assignments, collects full work logs, inspects diffs and quality checks, and can send correction assignments when results are sloppy.

OpenAI Evals is an open-source framework for evaluating LLMs and LLM systems. It provides an existing registry of benchmark/evaluation definitions (often data-driven), tooling to run local evaluations, and guidance for creating custom evals (including model-graded evals via YAML/templates).

LLM Workflow Engine (LWE) is a Python-based CLI and workflow manager for building and running LLM interactions (chat/tool use) from the shell, with a plugin architecture and support for multiple LLM providers (including OpenAI via the ChatGPT API).

Semantic Router (vLLM Semantic Router) provides a system-level routing layer for mixture-of-models by classifying or scoring incoming requests/contexts and dispatching them to appropriate models or pipelines.

claude-youtube is a Claude Code skill that provides YouTube creator-focused command workflows (channel audits, video SEO, scripts, hooks, thumbnails briefs, strategy, calendars, Shorts packages, analytics interpretation, competitor analysis, upload metadata, and ideation). It can operate with manual input, and optionally fetch live channel data via YouTube Data API (API key and OAuth2 for private analytics) and keyword/trend data via optional MCP integrations (e.g., DataForSEO, NanoBanana).

Provides a Claude Code “skill” (and bundled Python scripts) that performs SEO audits and GEO (AI search optimization) for a given site/URL. It fetches and analyzes pages (technical SEO, on-page, schema, links, local/international SEO, etc.), generates a scored audit plus a prioritized plan, and can output deployable fixes such as JSON-LD and meta/redirect/robots changes. It also includes GEO-specific checks like AI crawler allow/block status and llms.txt/RSL-related templates.

AI-Infra-Guard (A.I.G) is an AI red-teaming and security assessment platform that runs multiple scanners and evaluations, including OpenClaw security scanning, multi-agent workflow security scanning, MCP server/agent-skills scanning, AI infrastructure/component vulnerability scanning, and LLM jailbreak/prompt security evaluations. It exposes a web UI and a documented set of task-creation APIs (Swagger/docs) for running scans and retrieving results.

Khoj is a self-hostable “AI second brain” that provides chat and research-style Q&A over local and web/document sources using semantic search and LLMs. It supports multiple frontends (web, desktop/editor integrations) and can run agents and automations.

MetaClaw is a local proxy/agent runtime that sits in front of a user's personal agent (e.g., OpenClaw/CoPaw/etc.) and an OpenAI-compatible LLM API. It injects “skills” into prompts, can summarize sessions into new skills, and (optionally) runs asynchronous RL-style training (e.g., GRPO) on accumulated interaction data. It also includes a scheduler mode (madmax) to defer weight updates to idle/sleep/meeting windows and provides an Anthropic-compatible /v1/messages endpoint for Anthropic-native clients. A CLI (metaclaw) manages setup, config.yaml, starting/stopping, and daemon logging.

A Python “agent skill” that helps an AI coding agent plan, generate, and operationalize digital marketing campaigns. It includes utilities for UTM parameter generation/validation/auditing and brand-voice/compliance checking, plus reference materials for content strategy, email/social/SEO/GEO, and analytics/measurement frameworks.

ASK (Agent Skills Kit) is a Go-based CLI (and optional web/desktop UI) for managing “agent skills” from multiple sources. It installs/uninstalls skills, syncs them across multiple agent products (e.g., Claude, Cursor, Codex, Copilot, Windsurf, Gemini CLI, OpenClaw), supports version locking via an ask.lock file, and provides a security scanning/audit workflow plus offline/private-repo support.

Incipit is a Go-based CLI that takes structured resume data (JSON/Markdown) and generates formatted outputs (PDF via HTML/Chromium and/or LaTeX, DOCX, and Markdown), using a template system. It also provides AI commands to create/review/optimize resumes via multiple LLM providers (Anthropic, OpenAI, Google, or local Ollama), and can validate inputs using an exported JSON Schema.

A Claude Code skill that generates responsive, cross-client HTML email templates by producing MJML (MJML 4.x) and compiling it to production-ready, minified HTML suitable for Outlook and Gmail, including Outlook-specific handling (VML/ghost tables), Gmail clipping prevention, and accessibility-oriented MJML structuring.

A Claude Code “skill” that guides systematic genealogy research using Genealogical Proof Standard (GPS): analyzes historical documents (including handwritten, multi-language), helps manage an Obsidian vault (People/Places/Documents/Events, with cross-references and PROCESS.md), recommends sources/databases by time/region, handles naming/language variations, and supports GEDCOM import/export.

GLaDOS is an on-device (Python) voice assistant/agent framework that combines speech recognition, voice activity detection, text-to-speech, vision processing, an LLM core, and an MCP-based tool system to enable proactive/autonomous behavior (e.g., responding to camera/audio/time events) with long-term memory and configurable LLM backends (e.g., Ollama or OpenAI-compatible APIs).

gpt-oss is a Python repository providing reference inference implementations and tool/client examples for OpenAI’s open-weight gpt-oss models (gpt-oss-20b and gpt-oss-120b). It includes local inference via PyTorch, optimized (reference) Triton, and Apple Silicon Metal (reference), plus “harmony” response-format tooling and reference implementations of model tools (browser and python) and a sample Responses-API-compatible server.

HiClaw is an open-source Collaborative Multi-Agent OS that coordinates multiple agent “Workers” via a “Manager” agent, using Matrix rooms for transparent human-in-the-loop task coordination. It includes a local/self-hosted Matrix server (Element Web + Matrix backend), a MinIO-backed shared filesystem for inter-agent exchange, and a Higress AI Gateway for proxying/isolating real credentials (workers use consumer tokens).

Developer Kit for Claude Code: a modular marketplace of plugins (core + language/infra/tooling plugins) that provide reusable skills, agents, and slash commands to drive a repeatable development workflow (brainstorm/spec-to-tasks/task management/feature development/code review & debugging).

Neuralangelo is an official implementation for high-fidelity neural surface reconstruction, built on the NVIDIA Imaginaire library. It provides training and inference (including isosurface/mesh extraction) for reconstructing a 3D surface from images/video with known camera poses.