Report by The Next New Thing.
Scroll for the report ↓
The official list of community-built plugins for Claude Code and Claude Cowork. Two commands install any of them — and every one on the list passed a security scan before it landed.
View on GitHub →claude plugin marketplace add anthropics/claude-plugins-community, then claude plugin install <name>@claude-community.
Submissions go through clau.de/plugin-directory-submission, pass automated security scanning, and are approved before they appear.
marketplace.json syncs nightly from Anthropic’s internal review pipeline. Pull requests opened against the repo are closed automatically.
Public repository for Agent Skills
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
A collection of notebooks/recipes showcasing some fun and effective ways of using Claude.
Somebody took hundreds of great AI images, worked backwards to the prompt that made each one, and rewrote them all as reusable templates instead of paragraphs you retype.
View on GitHub →Each one reverse-engineered from a finished image and rewritten as a reusable structure, not copied as a paragraph of prose.
Subject, lighting, materials, layout and visual detail are split into composable parts, so you swap one element without rewriting the whole prompt.
gpt-image2.canghe.ai opens large previews, copies full prompts, filters by style or scenario and links back to the source case on GitHub.
CodexGuide:面向全球初学者、创作者、开发者与团队的 Codex 实践指南
AI 漫剧制作 APP,能一键生成剧本、分镜及合成视频,让你在手机上也能快速制作漫剧。
让 OpenClaw稳定的连上你的个人微信
A ready-made Arch Linux setup from the creator of Ruby on Rails. One command turns a bare install into a fully themed, keyboard-driven desktop where every app and config is already chosen for you.
View on GitHub →Window manager, terminal, editor, theme and hotkeys are decided for you — the way Rails picks defaults so you do not have to.
manual/ is its authoritative source and mirrors to learn.omacom.io, covering everything from coming off a Mac to dual boot and unattended installs.
Ten days after the 4.0 milestone — and the repo took another push the morning this was built.
Imagine giving your AI helper a big key ring. One connection lets it use thousands of apps, such as Gmail and Notion, to do everyday chores for you.
zapier.com/mcpIf you write software, this is the same key ring handed to your program instead of your chatbot — a few lines of code and your app can use those thousands of apps too.
zapier.com/sdkLogitech’s Options+ app is heavy, wants an account, and doesn’t run on Linux. This is a Rust rewrite that talks to the mouse directly — no account, no telemetry.
View on GitHub →Options+ never shipped for Linux. OpenLogi runs on macOS, Linux and Windows from the same codebase.
Every binding lives in plain text you can sync between machines however you like, and there is a real CLI alongside the GUI.
Zoom, focus, exposure and white balance are written straight to any Logitech UVC camera, so the change applies in Zoom, Meet and OBS at once.
🌊 AChat - An open-source/self-hosted/local-first AI platform, designed for enterprises and teams, perfectly combining powerful local processing capabilities with seamless remote synchronization.
Claude Code source code recovered from source map
Turn AI subscriptions into standard API endpoints. OpenAI & Anthropic compatible gateway — for Amp Code, Cursor, Factory CLI, and more.
A coding agent that runs on your own machine from the command line. Install it, sign in with the ChatGPT account you already pay for, and it works on your codebase in your terminal.
View on GitHub →Sign in with ChatGPT on Plus, Pro, Business, Edu or Enterprise. An API key works too, but it needs extra setup.
A curl script on Mac and Linux, a PowerShell line on Windows, or npm install -g @openai/codex / brew install --cask codex.
The same agent runs in your shell, inside VS Code, Cursor or Windsurf, as a desktop app via codex app, and as Codex Web at chatgpt.com/codex.
Robust Speech Recognition via Large-Scale Weak Supervision
Examples and guides for using the OpenAI API
A toolkit for developing and comparing reinforcement learning algorithms.
A desktop app for running AI agents on your own machine, where every model message, tool call and permission decision is written to an append-only log — so you can always go back and see what the agent did.
View on GitHub →Model messages, tool calls, tool results and how a turn ended are all written down. The UI is a view of that record, not the only copy.
Maka can drop old tool output from the next prompt without throwing away the saved evidence, so trimming context never destroys the trail.
Desktop (Electron + React), a TUI/CLI (maka, maka run) and a benchmark harness (maka eval run) all go through the same Runtime Host.
Apache Superset is a Data Visualization and Data Exploration Platform
Apache ECharts is a powerful, interactive charting and data visualization library for browser
Apache Airflow - A platform to programmatically author, schedule, and monitor workflows
Mojo is a programming language built to make AI code run fast without dropping to C++. Version 1.0 means the team is now promising parts of it won’t change under you.
View on GitHub →Standard-library APIs are now being marked stable, starting with a small set that grows through the 1.x line. Stable means they won’t break.
1.0 carries more breaking changes than usual, but nearly every one ships with a deprecated alias and a compiler fix-it.
std.gpu.*, std.algorithm and std.benchmark are now max.*, forming a new MAX accelerator library alongside the existing AI kernels.
A tool for working with stacked PRs on github.
Everything you need to know about LLM inference
Learn GPU Programming in Mojo🔥 by Solving Puzzles
Give it a subject or a keyword. It writes the script, finds the footage, generates the voiceover, burns in subtitles, adds music, and hands you a finished HD short — then posts it for you.
View on GitHub →Script, matched footage, voiceover, subtitles and background music, rendered to 9:16 1080×1920 or 16:9 1920×1080, in batches so you can pick the best take.
One-click cross-posting to TikTok, Instagram and YouTube Shorts once the video renders.
Free HD clips from Pexels, Pixabay and Coverr, your own files, or brand-new visuals generated from your script’s keywords by WaveSpeed AI.
快速提取音视频内容,整理成一份结构化的markdown笔记
Safety-first disk cleaner and disk space analyzer for macOS and Windows, with duplicate file finder, large file cleanup, app uninstaller, and startup manager
本地优先的 macOS 与 Windows 语音输入工具,支持端侧 SenseVoice 识别、全局语音输入、文件转写及可选 AI 校对。Local-first voice dictation for macOS and Windows with on-device SenseVoice, typing into any app, file transcription, and optional AI proofreading.
Every AI company gives away a small free tier. Alone each one is a toy. This bolts 34 of them together behind a single endpoint and switches providers automatically when one runs out.
View on GitHub →Point any OpenAI-compatible client at your local server and it routes across every provider you’ve added a key for. 474 model families in the catalog.
The router picks the best available model, moves on when one rate-limits, and tracks usage per key so you stay under every free-tier cap.
A signed feed delivers new models and quota changes without a git pull, because the free-tier landscape shifts weekly.
Self-hosted team chat where humans and agents are first-class members — channels, DMs, threads, reactions, file uploads, and an agent runtime.
Self-hosted personal AI assistant with MCP-native, bio-inspired memory. OpenClaw-compatible skills. LoCoMo F1 0.48 vs 0.38.
CLI tool for AI agents to register domain names for free
A personal finance app you run yourself. It pulls in your real bank transactions, categorises them, and shows your net worth — and none of it leaves a machine you control.
View on GitHub →curl -fsSL https://usesecuro.com/install.sh | bash installs Docker if you don’t have it. Windows runs docker compose up --build.
Enable Banking covers ~2,500 European PSD2 banks, SimpleFIN covers US and international, Pluggy covers Brazil. Plus OFX, QIF, CAMT and CSV import.
A self-hosted LLM chat with tool-use across your transactions, plus a per-agent RAG knowledge base — running on your infrastructure, not a vendor’s.
A self-improvement skill for AI coding assistants. At the end of a session it reads back the tasks, corrections and dead ends, then keeps only the lessons that would change how future work gets handled.
View on GitHub →I built Hindsight as a self-improvement skill for AI coding assistants. At the end of a session it reviews the tasks, corrections, and dead ends, then keeps only the lessons that would change how future work gets handled.
Building something? Send it to hi@TheNextNewThing.ai
A menu bar app showing what’s left of your Claude and ChatGPT subscriptions, broken down by project, session and task, so you can see what’s burning through the usage.
View on GitHub →I watch nearly every episode, and ModelDeck is a menu bar app that cleanly shows how much of your Claude, ChatGPT, and soon Grok subscriptions you have left, plus a breakdown by project, session, and task so you can see what’s burning through usage and stretch it further.
Building something? Send it to hi@TheNextNewThing.ai
A memory system and codebase analyzer built to make long agent sessions more reliable, with an integration for people already using graphify.
View on GitHub →I finally have this AI memory system and codebase analyzer to a point it’s legitimately helping my sessions be more successful, efficient, and reliable. It has a little integration with graphify for people using that as well.
Building something? Send it to hi@TheNextNewThing.ai
Local-first project memory for coding agents. It indexes repository evidence, preserves decisions and checkpoints with provenance, and prepares focused context for the next session.
View on GitHub →I follow your GitHub repo breakdowns, especially the episodes on local AI tools and practical agent workflows. Rta-Smriti Brain is an MIT-licensed, local-first project memory for coding agents. It indexes repository evidence, preserves decisions and checkpoints with provenance, and prepares focused context so a new session doesn’t have to reconstruct the project again.
Building something? Send it to hi@TheNextNewThing.ai
Pulls your subscriptions into a local staging folder and turns them into structured Markdown — full descriptions, transcripts and clickable timestamps — then lets you search locally instead of spending tokens. The first ingest did 11,500 videos from 41 channels in ten minutes.
View on GitHub →This is my first GitHub submission. I’m new to the AI space — I learned GitHub a couple of weeks ago and spent about four weeks immersed in tutorials before building this to solve my own struggle with managing YouTube subscriptions. It uses Andrej Karpathy’s LLM Wiki framework: raw data into a local staging folder, processed into structured Markdown, with a prompt asking whether you want a local search to cut the token overhead.
Building something? Send it to hi@TheNextNewThing.ai
A client-installable agent skill, not a SaaS dashboard. /reimagine-it reads the source file and redesigns from its own nouns, dates and colors.
You asked publicly for GitHub repos and creators worth featuring. This is a client-installable agent skill, not a SaaS dashboard: /reimagine-it reads the source file and redesigns from its own nouns, dates, and colors instead of inventing a vibe.
Building something? Send it to hi@TheNextNewThing.ai
A native Mac app where several agents share one conversation and consult each other mid-task, backed by a server you can reach from your phone. Signs in with subscriptions instead of API keys, and runs local models with no key at all.
View on GitHub →Radiant is a native Mac app where multiple AI agents share one conversation and consult each other mid-task, backed by a server you can reach from your phone and other Macs. It signs in with subscriptions instead of API keys, runs local models with no key at all, and connects to Hermes and OpenClaw agents already on your machine.
Building something? Send it to hi@TheNextNewThing.ai
A free, fully local voice-intelligence app for macOS — dictation, meeting capture, transcription and voice creation in one tool. Unlike subscription dictation apps, it needs no account and keeps speech and transcripts on your Mac.
View on GitHub →Kiki is a free, fully local voice-intelligence app for macOS that combines system-wide dictation, meeting capture, recording transcription, and voice creation in one private tool. Unlike subscription dictation apps such as Wispr Flow, it requires no account and keeps speech and transcripts on your Mac.
Building something? Send it to hi@TheNextNewThing.ai
A rubric that scores an agent across seven dimensions, twice: once while it’s engaged with you (Active), once across idle periods when nothing triggered it (Ambient). The two are reported side by side and never averaged — the spread is the point.
View on GitHub →Everyone in the agent space says “autonomous” and nobody defines it. AAS scores a system on a 0–5 scale twice — Active and Ambient — and reports them side by side. Claude Code comes out at 2.29 active and 0.57 ambient; Manus at 2.43 and 0.71; Siri at 1.71 and 0.29. The tools that feel the most agentic in use are the ones that go closest to zero the moment you stop typing. It’s designed to be rated independently — if you score something and land somewhere different than I did, I want to hear it.
Building something? Send it to hi@TheNextNewThing.ai
New hot repos every week — don't miss the next drop.
It tells YouTube to show this to more builders like you.
Tell me which repo you'll try first — or what to cover next.
New hot repos every week. Subscribe so you never miss the next drop.