GitHub Trending Weekly Digest — May 11-16, 2026

This week's GitHub Trending landscape (May 11-16) reveals a powerful shift toward AI agent infrastructure, persistent memory systems, and local-first tooling. We tracked 14 standout projects across 6 days, with 4 persistent dominators, 7 recurring contenders, and 3 emerging stars.

🔥 Persistent Dominators (3+ Days)

These projects consistently held trending positions throughout the week, demonstrating sustained community interest.

🏆 OpenHuman

🔗 github.com/tinyhumansai/openhuman
📅 Trending: May 13, 14, 15

What it does:
Private, simple, and incredibly powerful personal AI super-intelligence assistant. Integrates 118+ third-party services (Gmail, Notion, GitHub, Slack, Stripe, Calendar) with auto-fetch every 20 minutes. Memory Tree + Obsidian-compatible vault compresses everything into local Markdown files. Desktop mascot that talks, lip-syncs, and joins Google Meet as a real participant.

Why it matters:
Solves the cold-start problem: most AI assistants need days/weeks to learn your work context. OpenHuman syncs your data in 20 minutes and builds complete (compressed) context of your inbox, calendar, repos, docs, and messages. All data stays local, encrypted. Built-in web search, scraper, full code toolset (filesystem, git, lint, test, grep), and native voice I/O.

Tech: Rust, Node.js 24+, Tauri, Memory Tree (SQLite), TokenJuice compression, Ollama local AI, ElevenLabs TTS

🏆 Superpowers

🔗 github.com/obra/superpowers
📅 Trending: May 13, 14, 16

What it does:
A complete software development methodology and agent skills framework. Structured workflow: brainstorming → design approval → planning → subagent-driven development → test-driven development (RED-GREEN-REFACTOR) → code review → merge. Enforces engineering best practices so agents can work autonomously for hours.

Why it matters:
Coding agents tend to jump straight into writing code without design, tests, or reviews. Superpowers introduces industrial-grade workflows that keep agents on track. Available on Claude Code official plugin marketplace and compatible with Codex CLI/App, Gemini CLI, OpenCode, Cursor, GitHub Copilot CLI.

Tech: Agent Skills standard, TDD, Git worktrees, systematic debugging, parallel agents

🏆 Scientific Agent Skills

🔗 github.com/K-Dense-AI/scientific-agent-skills
📅 Trending: May 14, 15, 16

What it does:
135 ready-to-use agent skills for research, science, engineering, analytics, finance, and writing. Covers bioinformatics, genomics, cheminformatics, drug discovery, proteomics, clinical research, precision medicine, medical AI, imaging, ML, materials science, physics, astronomy, engineering, data analysis, geospatial science, lab automation, scientific communication, multi-omics, and protein engineering.

Why it matters:
Transforms AI agents into "AI Scientists" on your desktop. While agents can use any Python package or API, these curated skills provide documentation and examples that make them significantly more powerful and reliable in scientific workflows. Supports Agent Skills standard for Cursor, Claude Code, Codex, Gemini CLI, etc. Install with one command: npx skills add K-Dense-AI/scientific-agent-skills

Tech: 78+ public databases (PubChem, ChEMBL, UniProt, COSMIC, ClinicalTrials.gov, etc.), 70+ optimized Python packages (RDKit, Scanpy, PyTorch Lightning, BioPython, etc.), 9 scientific integrations (Benchling, DNAnexus, LatchBio, OMERO, Protocols.io), 30+ analysis/communication tools

🏆 AiToEarn

🔗 github.com/yikart/AiToEarn
📅 Trending: May 11, 12, 13

What it does:
AI-powered content creator monetization platform. One-click distribution to 10+ platforms (TikTok, YouTube, Douyin, Xiaohongshu, Instagram, Twitter, Facebook, LinkedIn, Threads, Pinterest), content marketplace, AI auto-reply, automated engagement.

Why it matters:
Solves the multi-platform publishing nightmare for creators. Agent-driven workflow: Monetize (CPS/CPE/CPM marketplace) → Publish (batch distribution) → Engage (AI comment replies) → Create (auto video generation). Supports 5 integration modes including MCP for Claude/Cursor.

Tech: TypeScript, Node.js, AI video generation (Grok, Veo, Seedance), MCP protocol, OpenClaw plugin

🔄 Recurring Contenders (2 Days)

These projects made multiple appearances, showing strong momentum.

ByteDance UI-TARS Desktop

🔗 github.com/bytedance/UI-TARS-desktop
📅 Trending: May 11, 12

What it does: Multimodal AI Agent tech stack connecting frontier AI models with agent infrastructure. Enables natural language control of browsers, desktop apps, and terminals through hybrid GUI + DOM strategies.

Why it matters: Bridges the gap between cutting-edge multimodal LLMs and real-world automation. One-click CLI with out-of-the-box browser and desktop control — no complex scripting required.

Tech: TypeScript, Node.js, Electron, Playwright, MCP Protocol

CloakBrowser

🔗 github.com/CloakHQ/CloakBrowser
📅 Trending: May 11, 12

What it does: Source-level stealth Chromium browser that passes 100% of anti-bot detection tests. Drop-in replacement for Playwright/Puppeteer with 57 C++ patches to canvas, WebGL, audio, fonts, GPU fingerprints.

Why it matters: Unlike config-level patches (playwright-stealth) that break with every Chrome update, CloakBrowser modifies Chromium at source level. Achieves reCAPTCHA v3 score 0.9 (human-level) and passes Cloudflare Turnstile, FingerprintJS, BrowserScan.

Tech: Python, JavaScript, C++ (Chromium source mods), ONNX

Easy-Vibe

🔗 github.com/datawhalechina/easy-vibe
📅 Trending: May 11, 12

What it does: Complete Vibe Coding tutorial for absolute beginners. Teaches how to go from idea to deployed product in the AI era through 3 stages: prototype building, full-stack development (Figma/Supabase/Stripe), and advanced topics (Claude Code/MCP/Skills/Agent Teams/cross-platform apps).

Why it matters: In 2026, programming starts with describing needs. Easy-Vibe provides a systematic learning path for the AI-native generation, covering everything from AI IDE tools to multi-platform deployment (WeChat Mini Programs, Android/iOS, PWA, browser extensions, Electron). Real user stories from rural teachers, college students, and truck drivers prove accessibility.

Tech: JavaScript, Node.js, React, Claude Code, MCP, Figma, Supabase, Stripe

SuperSplat

🔗 github.com/playcanvas/supersplat
📅 Trending: May 11, 12

What it does: Free open-source 3D Gaussian Splat editor that runs entirely in the browser. No download, no installation — just open superspl.at/editor and start editing, inspecting, optimizing, and publishing 3D Gaussian Splats.

Why it matters: Democratizes access to cutting-edge 3D Gaussian Splatting technology. Web-first approach eliminates installation friction and enables cross-platform collaboration. Perfect for newcomers to explore next-gen 3D reconstruction without heavy tooling.

Tech: TypeScript, WebGL, PlayCanvas Engine

AgentMemory

🔗 github.com/rohitg00/agentmemory
📅 Trending: May 13, 14

What it does: #1 AI coding agent persistent memory system based on real benchmarks. Solves the repeat-yourself-every-session problem. Built on iii engine with BM25 + Vector + Knowledge Graph (RRF fusion), achieving 95.2% R@5 on LongMemEval-S benchmark. Supports 104 REST APIs, MCP protocol, and 12 Claude Code hooks.

Why it matters: Every AI coding assistant forgets everything after each session. AgentMemory captures agent behavior in background, compresses into searchable memories, and injects right context at next session start. Saves ~170K tokens/year ($10) vs LLM summarization (~650K tokens, $500). Works with Claude Code, OpenClaw, Cursor, Gemini CLI, OpenCode, Codex CLI, Cline, Goose, and more via MCP.

Tech: iii engine, SQLite (no Qdrant/pgvector), MCP, REST API (104 endpoints), BM25/Vector/KG hybrid retrieval

RuView

🔗 github.com/ruvnet/RuView
📅 Trending: May 14, 15

What it does: Turns ordinary WiFi signals into spatial intelligence. Detects people through walls, measures breath and heart rate, tracks movement — no cameras, no wearables, pure physics. Uses CSI (Channel State Information) analysis to achieve 17-joint pose estimation (COCO format), vital signs monitoring (breath 6-30 BPM, heart rate 40-120 BPM), through-wall detection (up to 5m), and multi-person tracking.

Why it matters: Solves privacy and accessibility issues of camera-based detection and wearable dependency. Use cases: elderly care (fall detection, sleep monitoring), hospital wards (breath/heart rate monitoring), retail (foot traffic), office space utilization, smart homes (through-wall presence), fitness (posture detection), search & rescue (life sign detection through debris).

Tech: ESP32-S3 ($9) + Cognitum Seed ($140) or 3-6 ESP32-S3 mesh, CSI multi-band fusion (6 WiFi channels × 56 subcarriers), RuVector (attention, graph algorithms, compression, field models), WiFlow neural network, 65 WASM modules

Supertonic

🔗 github.com/supertone-inc/supertonic
📅 Trending: May 15, 16

What it does: Blazing fast local multilingual TTS (text-to-speech) via ONNX. Supports 31 languages, fully offline, privacy-safe. 99M parameters (much smaller than 0.7B-2B open-source TTS), runs on CPU (faster than large models on A100 GPU), outputs 44.1kHz high-quality audio. Can run on Raspberry Pi, e-readers, browsers, iOS devices.

Why it matters: Cloud TTS APIs (ElevenLabs, OpenAI, Google) require internet, have latency, pose privacy risks, and cost money. Supertonic provides a fully local solution: fast on CPU, supports 31 languages (Chinese, Japanese, Korean, Arabic, European languages, etc.), low latency (RTF 0.3× on Raspberry Pi). Perfect for e-book readers, accessibility devices, edge devices requiring privacy protection or offline operation.

Tech: ONNX Runtime (cross-platform inference), onnxruntime-web (browser WebGPU), Speech autoencoder + Flow-matching TTS + LARoPE (length-aware rotary position encoding)

🌱 Emerging Stars (1 Day)

Fresh projects that caught the community's eye.

Bun

🔗 github.com/oven-sh/bun
📅 Trending: May 16

What it does: All-in-one ultra-fast JavaScript runtime, bundler, test runner, and package manager. Written in Zig, powered by JavaScriptCore. Dramatically faster startup and lower memory usage than Node.js. Integrates build, test, and package management — replace thousands of node_modules dev tools with one binary.

Why it matters: Addresses Node.js performance bottlenecks. Bun redefines the JavaScript ecosystem's performance baseline. Startup speed and memory footprint significantly better than Node.js. One binary for everything developers need (runtime, bundler, tester, package manager) — simplifies tooling complexity.

Tech: Zig + JavaScriptCore, Node.js API compatible, npm ecosystem compatible, built-in bundler (esbuild-like), test runner, package manager

Open Generative AI

🔗 github.com/Anil-matcha/Open-Generative-AI
📅 Trending: May 16

What it does: Open-source AI image/video generation studio with 200+ models, zero content censorship. Integrates Flux, Midjourney, Kling, Sora, Veo and more. Supports text-to-image, image-to-image, text-to-video, image-to-video, lip sync. Up to 14 reference images (multi-image models). Fully local or self-hosted, no content filters.

Why it matters: Commercial AI video platforms (Midjourney, Runway) have content filters, subscription fees, and no self-deployment. Open Generative AI provides an open-source alternative integrating 200+ models with full local control and no censorship. Perfect for creators who need freedom and control.

Tech: Next.js 14 + React 18 + Tailwind CSS, Muapi.ai (unified model gateway), sd.cpp (bundled: CPU/Metal/CUDA/Vulkan/ROCm, SD 1.5/SDXL/Z-Image), Wan2GP (BYO server: PyTorch + CUDA/ROCm, Flux/Qwen-Image/Wan 2.2/Hunyuan Video/LTX Video), Electron (macOS/Windows/Linux)

Telegraf

🔗 github.com/influxdata/telegraf
📅 Trending: May 13

What it does: Agent for collecting, processing, aggregating, and writing metrics, logs, and arbitrary data. 300+ plugins covering device monitoring (OPC UA, Modbus), log collection (File, Tail), messaging systems (AMQP, Kafka, MQTT), monitoring protocols (OpenTelemetry, Prometheus), network devices (Cisco TelemetryMDT, gNMI), system monitoring (CPU, Memory, Disk, Docker, Nvidia SMI), and generic interfaces (Exec, HTTP, SNMP, SQL).

Why it matters: Infrastructure monitoring never goes out of style. Telegraf remains a solid choice for multi-source data collection (IoT, cloud-native, IT infrastructure, industrial data, logs, message queues) with a plugin-based architecture that keeps complexity low. Official project from InfluxData with 1,200+ contributors.

Tech: Go, TOML config, 300+ plugins, standalone binary with no external dependencies

📊 Weekly Themes

The May 11-16 trending landscape reveals four dominant themes:

AI Agent Infrastructure Maturity

OpenHuman, Superpowers, AgentMemory, Scientific Agent Skills: The ecosystem is moving beyond simple chatbots toward agents with persistent memory, structured workflows, and specialized domain skills
Memory is king: Both OpenHuman (Memory Tree) and AgentMemory (hybrid retrieval) solve the cold-start problem with local-first persistent memory
Standardization emerges: Agent Skills standard (agentskills.io) gaining traction across Cursor, Claude Code, Codex, Gemini CLI

Local-First Privacy Revolution

OpenHuman, Supertonic, Open Generative AI, Bun: Growing demand for tools that run entirely on your machine
Privacy by architecture: OpenHuman stores all data locally encrypted; Supertonic eliminates cloud TTS; Open Generative AI bypasses platform censorship
Performance wins: Local execution often outperforms cloud (Supertonic faster than cloud TTS on Raspberry Pi; Bun faster than Node.js)

Vertical Domain Specialization

Scientific Agent Skills (135 skills for research), AiToEarn (creator monetization), RuView (spatial intelligence via WiFi)
Move from general-purpose AI to domain-specific tooling with deep integration (databases, specialized Python packages, industry workflows)

Developer Experience Innovation

Superpowers (enforced TDD + Git worktrees), Easy-Vibe (Vibe Coding for beginners), Bun (all-in-one tooling)
Tooling that codifies best practices (Superpowers' RED-GREEN-REFACTOR loop) or drastically simplifies onboarding (Easy-Vibe's beginner path)

✨ Key Takeaways

Memory is the new frontier — AI agents without persistent memory are like developers without RAM. OpenHuman and AgentMemory show the path forward.
Local-first isn't just privacy — it's performance, cost savings, and control. From Supertonic's offline TTS to OpenHuman's compressed Memory Tree, staying local wins.
Agent skills are becoming standardized — The Agent Skills standard (agentskills.io) is emerging as the "npm for agent capabilities," enabling cross-platform skill sharing.
Vertical beats horizontal — General-purpose AI is commoditizing. The value moves to domain-specific integrations (Scientific Agent Skills' 78 databases + 70 Python packages, AiToEarn's 10-platform distribution).
Structured workflows > raw prompts — Superpowers proves that enforcing engineering discipline (TDD, code review, Git worktrees) lets agents work autonomously for hours without derailing.

Compiled by Tommy Zhang | May 17, 2026