← All transcripts

Top Open-Source GitHub Projects : open-code-review, LobeHub, hyperframes, devvm & graphicle #309 Transcript, AI Summary & Key Points

ManuAGI - AutoGPT Tutorials · 2 days ago · Science & Technology · 15:46 · EN-US

Watch on YouTube

AI Summary

Twenty trending open-source GitHub projects for developers this week, from AI code review to agent workflows. Open Code Review is Alibaba's AI code review CLI pairing deterministic pipelines with an LLM agent that returns line-level comments and auto-applies fixes. Plannotator gives AI coding agents a local visual review surface for plans, diffs and artifacts. Claude Trading Skills turns Claude into a systematic research assistant for traders. Chrome DevTools MCP connects coding agents to a live Chrome for real debugging. Auto-ontology is NVIDIA's semantic layer grounding agent queries in enterprise data with certified text-to-SQL. LobeHub is a free self-hostable AI agent workspace that operates a whole team of agents. Hyperframes renders HTML frames into video, built for agent authoring. Anthropic ships knowledge-work plugins for Claude, and Google Chrome ships modern-web-guidance with over 100 expert-verified guides. Exact is Expo's experimental cross-platform native app runtime designed for agents. Supermimic trains humanoid whole-body control with one differentiable world model on a single GPU. Henge runs Mac OS on iPhone and iPad via its own JIT emulator. MyGo Music is a native Go desktop music player. DevVM gives each project folder an isolated VM for sandboxed coding agents. CUI is a native C UI library of around 150 components near 240 KB. Pinky lets agents operate a real phone through MCP with approval gates. Chrome Capture records Chrome tabs to AV1 MP4 at high frame rates without FFmpeg. Phone Use views and controls an unmodified iPhone from Linux. Graphicle gives coding agents a code map and project memory as one Rust MCP server. Ding morphs a single toast notification in place with spring animations.

Key Points

  • Open Code Review — Alibaba's AI code review CLI reads your git diff while an LLM agent reads full files and searches the codebase; multi-language rules catch null pointer bugs, thread safety, XSS and SQL injection, and it auto-applies fixes, working with OpenAI and Anthropic plus plugins for Claude Code and Codex
  • Plannotator — a local review surface where an agent's plan opens in your browser for visual markup, then one click sends feedback back; it also reviews code diffs, PRs, markdown and HTML from Git, Jujutsu or Git Butler, saves approved plans to Obsidian, and works with Claude Code, Codex and Open Code
  • Claude Trading Skills — a collection of skills.md files covering market analysis, stock screening, technical charting, backtesting, economic calendars and 13F filings; journals a thesis from idea to close, US-focused, MIT licensed, and for research not financial advice
  • Chrome DevTools MCP — Google's official MCP server lets an AI agent drive real Chrome with Puppeteer, reading the console, inspecting network requests, taking screenshots and recording performance traces; runs with one npx command, plugs into Claude Code, Cursor, Copilot and Gemini, Apache licensed
  • Auto-ontology — NVIDIA's semantic layer builds a graph of database context from metadata and guides agents to generate deterministic certified text-to-SQL queries with provenance; ships a web app, REST API and MCP server, runs on Postgres via Docker, Apache licensed
  • LobeHub — a free self-hostable AI agent workspace, formerly Lobe Chat, that treats each agent as a unit of work and acts as a chief agent operator hiring, scheduling and reporting on a team running around the clock; supports many model providers, a plugin and skills marketplace, knowledge bases, and a one-click free deploy
  • Hyperframes — HeyGen's open-source framework makes video from HTML frames animated with GSAP, no After Effects or timeline app; an agent can author and revise every shot, bundled skills teach it engine timing and idioms, and it plugs into Claude Code and Cursor, MIT licensed
  • Knowledge Work Plugins — Anthropic's open-source plugin suite for Claude in Co-work and Claude Code, bundling skills, commands and settings for productivity, sales, marketing, finance, legal and data analysis, organized by team and installable individually

AI in practice

Used for

Agents

  • LobeHub agents — Run a whole team of agents around the clock, managed by a chief agent operator that hires, schedules, and reports on them. 1 held

Tools & resources

20 items

ANo. 7390
AIAINotes.us AI product

Auto-Ontology

Open source · NVIDIA/auto-ontology

Auto-Ontology is an open-source Python framework and service from NVIDIA that builds a semantic, graph-based context layer over enterprise databases for LLM and agent workflows. It ingests schemas, column profiles, sample values, and query history, then compiles business terms mapped to columns, reviewed SQL attributes, and inferred table relationships into a PostgreSQL-backed ontology with pgvector embeddings. People can inspect, edit, test, and certify the generated definitions through a web UI or API. An agent resolves question entities to ontology terms, retrieves candidate columns and SQL attributes, generates SQL, validates filter values and the query against the source database, repairs failed queries, and returns the answer with its SQL and execution steps. The system can check whether a question is answerable before querying and exposes read-only ontology and question-answering tools through MCP, agent skills, and a REST API. It supports PostgreSQL, MySQL, Databricks, Snowflake, ClickHouse, Trino, Apache Kyuubi, HeavyDB, DuckDB, and SQLite, and can run with Docker Compose. The repository is licensed under Apache-2.0.

Mentioned in
2 videos
Kind
AI
CNo. 8018
AIAINotes.us Tool

chrome-capture

Open source · paulirish/chrome-capture

chrome-capture is an open-source CLI and npm library for recording Chrome tab video and optional tab audio directly to AV1 MP4. It uses Chrome's Page.startScreenRecording API rather than the older Page.startScreencast pipeline, sending frames from the compositor to Chrome's encoder over shared memory instead of JPEG-encoding frames for CDP and finalizing them through FFmpeg. This enables recording at up to 120 fps and 2160p where the display and Chrome configuration support it, with near-instant stopping and the ability to start and stop recording around selected interactions. The CLI accepts a URL, output path, duration, frame-rate and size options, and can run headless or headful. The npm API's startRecording function accepts a Playwright Page, Puppeteer Page, or raw CDP session; it also provides a Playwright test fixture, a CdpScreenRecorder, configurable capture areas, optional audio, and output as either an MP4 file or a buffer.

Mentioned in
1 video
Kind
Other
CNo. 3309
AIAINotes.us AI product

Chrome DevTools for agents

Open source · ChromeDevTools/chrome-devtools-mcp

Chrome DevTools for agents (chrome-devtools-mcp) is an MCP server from the Chrome DevTools project that lets coding agents control and inspect a live Google Chrome or Chrome for Testing browser. It uses Puppeteer for browser automation, including navigation and actions that wait for results, and exposes DevTools capabilities for recording and inspecting performance traces and insights, analyzing network requests, taking screenshots, and inspecting browser console messages with source-mapped stack traces. A CLI is also provided for use without MCP, with configuration options including headless, isolated, slim, concurrent-session, persistent-profile, WebSocket, and Android-debugging modes. It requires Node.js LTS, a current stable Chrome version or newer, and npm. Browser contents are exposed to MCP clients; performance tools may query the Google CrUX API, and usage statistics and update checks are enabled by default with an opt-out flag.

TypeScript
Stars
★ 57,903
Forks
4,861
CNo. 7920
AIAINotes.us AI product

Claude Trading Skills

Open source · tradermonty/claude-trading-skills

Claude Skills-based trading workflow toolkit for time-constrained individual investors, developed as an open-source personal project. It provides installable skills and workflow manifests for market-regime review, portfolio analysis, screening, technical chart analysis, position sizing, trade planning, journaling, post-trade review, and strategy research. Skills can use local calculations and files, public CSVs, chart screenshots, web search, or optional integrations such as Financial Modeling Prep, FINVIZ, Alpaca, yfinance, and CoinGecko; workflows keep human decision gates central and do not place orders or provide an automated signal service. The repository includes source skill folders, a canonical YAML skill index, workflow manifests, pre-built .skill packages for Claude's web app, documentation, validation scripts, and installation instructions for Claude Web and Claude Code. It is distributed under the MIT License and includes financial-risk and educational-use disclaimers.

Python
Stars
★ 6,679
Forks
1,386
CNo. 7677
AIAINotes.us Tool

CUI

Open source · yappologistic/CUI

CUI is a dependency-free native UI component library for C99, developed by yappologistic. It provides functions for drawing components and handling input, while applications describe their screen and state again each frame; named options configure component behavior, and components return interaction results. The library includes controls, dialogs, tables, data grids, charts, code and tool panels, planning views, canvases, and AI-oriented interface components. CUI targets Windows through Win32, macOS through Cocoa, and Linux or BSD through X11, with SDL2 and GLFW backends for other environments. It renders through OpenGL or a software fallback, and can use a custom renderer; the same library is also compiled to WebAssembly for browser-based live documentation. It is built with CMake, supports GCC, Clang, MSVC, and MinGW, and is distributed under the MIT license. The README notes that it has no accessibility tree yet, and native Wayland and mobile platform layers are not currently provided.

Mentioned in
2 videos
Kind
Other
DNo. 7660
AIAINotes.us AI product

devvm

Open source · devvmhq/devvm

devvm is a developer tool that creates an isolated Linux virtual machine for each project folder, using the project's files to detect and provision the required toolchains. It runs coding agents such as Claude Code, Codex, OpenCode, and Pi inside the VM, while preserving their settings, skills, MCP servers, and logins in a host-side store. The VMs run on Lima, mount the project at the same path as on the host by default, and use shared read-only toolchains and a reusable base image. New VMs use an allowlist network tier by default, while credentials such as GitHub and Claude Code tokens remain on the host. A clone mode gives the VM its own repository clone so changes can be reviewed before being fetched into the host repository. devvm is distributed for macOS and Linux, with installation through Homebrew or an install script, and is licensed under the MIT License.

Mentioned in
2 videos
Kind
AI
DNo. 7687
AIAINotes.us Tool

Ding

Open source · YusufCeng1z/ding

Ding is a zero-dependency TypeScript toast-notification library for React, Svelte 5, and plain DOM applications. It maintains one shared toast: each new call morphs the current toast instead of stacking another, while a spring solver animates its size, shape, and text and preserves momentum when animations are interrupted. Toasts support descriptions, avatars or thumbnails, primary and secondary actions, progress bars, countdowns, status types, promise-based updates, themes, six positions, right-to-left layouts, optional sounds, keyboard dismissal, live-region accessibility, and reduced-motion preferences. The project provides separate core, React, and Svelte packages, with styles injected from JavaScript, and is licensed under the MIT License.

Mentioned in
2 videos
Kind
Other
ENo. 8016
AIAINotes.us AI product

Exact

Open source · expo/exact

Exact is Expo's cross-platform application runtime for building one app that runs natively on the web, macOS, iOS, and Linux, with coding agents able to build, operate, inspect, and test it on each host. Developers describe views, state, actions, routes, and data shapes in Contract, a small declarative language, and implement data sources in TypeScript or Rust. The Contract compiler processes the source through syntax, type, analysis, and lowering passes into a compact plan; a runner executes that plan, using the browser's real DOM and CSS on the web, AppKit on macOS, UIKit on iOS, and a Rust layout kernel with GPU or CPU painting on Linux. The runtime has no webview or JavaScript bridge in the UI, and TypeScript is compiled to Hermes bytecode for native hosts and loaded after the first frame. Its agent driver provides ten cross-host operations, including tree inspection, screenshots, taps, typing, state, layout, logs, clock control, preferences, and performance checks; Contract test files run unchanged across hosts. Android and Windows support are early and incomplete. Exact is pre-1.0, changes frequently, and provides no API stability or backwards-compatibility guarantee. The workspace is MIT-licensed.

Mentioned in
1 video
Kind
AI
GNo. 6273
AIAINotes.us AI product

Graphicle

Open source · Graphicle-AI/graphicle

Graphicle is an open-source code graph and project-memory tool for coding agents, distributed as a single Rust program by KarLex AI. It parses repositories with tree-sitter to map files, functions, classes, calls, and imports; indexes project documents, decisions, procedures, and captured agent sessions; and returns cited excerpts with exact files and lines through its CLI and MCP server. Reviewed memories are linked to the functions they describe and become stale when those functions change, while synchronization keeps the graph current after edits and Git operations. The setup command connects supported MCP clients such as Claude Code, Cursor, Codex, VS Code, and Claude Desktop, and the app provides a browser interface for browsing sources and approving memories. It runs locally without an account, API key, telemetry, Python, Docker, or a database server for most features, stores data in SQLite, and is licensed under Apache-2.0.

Mentioned in
4 videos
Kind
AI
HNo. 7672
AIAINotes.us Tool

Henge

Open source · Leviidev/Henge

Henge is an open-source iOS and iPadOS app that emulates macOS on compatible iPhones and iPads. It runs the Mac's arm64 code through its own JIT engine, memory-management unit, and virtual devices rather than QEMU or Apple's Virtualization.framework. macOS graphics are rendered through a virtual GPU whose commands Henge replays with Metal; the device touchscreen acts as a trackpad, with keyboard and mouse support on iPad. Henge downloads Apple's macOS restore image and installs it directly on the device without requiring a Mac or modifying the image. It supports copy and paste between the guest and device, a shared folder exposed in Files and macOS, and external displays through compatible USB-C connections. The project distributes sideloadable IPA and TrollStore packages, requires iOS or iPadOS 17 or later on A14- or M1-class hardware or newer, and requires JIT. It is licensed under GPL-2.0-or-later and cannot be distributed through the App Store.

Mentioned in
2 videos
Kind
Other
HNo. 3524
AIAINotes.us AI product

HyperFrames

Open source · heygen-com/hyperframes

HyperFrames is an open-source framework for turning HTML, CSS, media, and seekable animations into deterministic MP4 videos. It can be used locally through a CLI, by AI coding agents through skills, or as a rendering core for hosted authoring workflows. Compositions are defined as HTML with data attributes for timing, tracks, media, and sub-compositions. The renderer seeks each frame in headless Chrome, uses animation adapters such as GSAP, CSS, Lottie, Three.js, Anime.js, or WAAPI, and encodes the result with FFmpeg; its toolchain includes preview, linting, inspection, local rendering, browser-based editing, reusable catalog components, and AWS Lambda rendering. The project requires Node.js 22 or newer and FFmpeg, does not require React or a build step, and is licensed under Apache 2.0. Its agent skills support workflows including product videos, explainers, pull-request videos, captions, motion graphics, slideshows, music-driven videos, and general compositions.

Mentioned in
6 videos
Kind
AI
KNo. 3979
AIAINotes.us AI product

Knowledge Work Plugins

Open source · anthropics/knowledge-work-plugins

Knowledge Work Plugins is an open-source repository of role-specific plugins for Claude Cowork and Claude Code. Each plugin bundles domain skills, MCP connector configuration, slash commands, and, where applicable, sub-agents for workflows such as productivity, sales, customer support, product management, marketing, legal, finance, data, enterprise search, and biomedical research. Plugins use file-based Markdown and JSON rather than application code, infrastructure, or build steps; users can install them through Cowork or the Claude Code plugin marketplace and customize their connectors, company context, and workflow instructions.

Mentioned in
5 videos
Kind
AI
LNo. 0985
AIAINotes.us AI product

LobeHub

Open source · lobehub/lobehub

LobeHub is a self-hostable platform for organizing teams of AI agents, presented as a chief agent operator. It treats agents as units of work and provides tools to create and configure agents, schedule and report on their operations, coordinate them in agent groups, and monitor shared workspaces and memory. The platform provides access to multiple models and modalities, an agent builder, a library of skills and MCP-compatible plugins, and an IM gateway for interacting with agents through existing chat services. It can be deployed using Docker or listed cloud deployment options, and is described as being under active development.

Mentioned in
3 videos
Kind
AI
MNo. 8015
AIAINotes.us AI product

Modern Web Guidance

Open source · GoogleChrome/modern-web-guidance

Modern Web Guidance is an open-source set of skills and a CLI for coding agents, stewarded by Google Chrome, Microsoft Edge, Mozilla Firefox, MDN, and the web development community. It supplies expert-curated guidance on modern browser APIs, CSS and responsive layouts, HTML and DOM, JavaScript, performance, accessibility, forms, security, passkeys, Web Components, built-in browser AI, WebAssembly, progressive web apps, and WebMCP, using browser compatibility data from the Baseline project to steer agents away from legacy patterns and toward progressive enhancements and lightweight fallbacks. The CLI installs the skills into supported coding-agent environments, searches guides locally with a zero-dependency embedding model and BM25 search engine, retrieves guides by ID, and scans codebases for legacy patterns. Guides include code patterns, compatibility caveats, and fallback guidance for the agent's context.

TypeScript
Stars
★ 7,716
Forks
682
MNo. 7674
AIAINotes.us Tool

MyGo Music

Open source · egoist/mygo-music

MyGo Music is a native desktop music player written in Go with the MyGo framework and no WebView. It plays local MP3, FLAC, Ogg Vorbis, WAV, ALAC, and AAC files with tags, covers, and LRC lyrics, and provides playlists, search, albums, artists, queue management, shuffle and repeat, loudness normalization, audio caching, session restore, and macOS-specific features such as a notch player, tray integration, and media keys. Additional services are supplied by JavaScript or TypeScript plugins. A plugin is a folder containing plugin.json and a script that can add accounts, catalogs, podcasts, stations, comments, custom browse pages, search, streaming, lyrics, playlists, and remote playback; plugins can be installed from GitHub, a local folder, or a ZIP archive. The player also supports podcast captions written on the device during playback using sherpa-onnx and Qwen3-ASR. It supports English and Simplified Chinese and is licensed under AGPL-3.0.

Mentioned in
2 videos
Kind
Other
ONo. 1274
AIAINotes.us AI product

OpenCodeReview

Open source · alibaba/open-code-review

Open Code Review is an open-source, AI-powered command-line code-review tool developed from Alibaba Group’s internal review assistant. It reads Git diffs and sends changed files to a configurable LLM through an agent with tool-use capabilities; the agent can read complete files, search the repository, and inspect other changed files for context before producing structured comments tied to source lines. Its hybrid architecture combines deterministic engineering with dynamic agent decisions: deterministic stages select and bundle files, match rules to file characteristics, and apply independent comment-positioning and reflection modules, while scenario-specific prompts and tools guide context retrieval and review. The `ocr scan` command audits complete files or directories without requiring a meaningful diff, and the project provides multilingual rules for issues including null-pointer exceptions, thread safety, cross-site scripting, and SQL injection. It supports configurable model endpoints, including OpenAI- and Anthropic-compatible endpoints, and requires Git 2.41 or newer.

Go
Stars
★ 53,853
Forks
4,437
PNo. 7664
AIAINotes.us Tool

phoneuse

Open source · lxe/phoneuse

phoneuse is a Linux tool for viewing and operating an unmodified iPhone without a jailbreak, developer account, cable, or phone-side app. It receives the iPhone's screen through AirPlay and sends mouse and keyboard input through Bluetooth HID, using either the Linux PC's Bluetooth adapter or an optional ESP32 bridge. A mirror window and local browser interface support direct interaction, while a CLI, MCP server, and Claude Code skill expose actions such as reading screen text, taking screenshots, tapping elements, typing, unlocking with approval, and navigating the phone. The project is an MIT-licensed technology demo under heavy construction, tested by its author on an iPhone 17 Pro with Arch Linux, Hyprland, and an NVIDIA GPU.

Mentioned in
2 videos
Kind
Other
PNo. 8017
AIAINotes.us AI product

pinky

Open source · lxe/pinky

Pinky is an open-source desktop application that lets AI agents such as Claude Code, Codex, and Cursor operate real iPhone and Android devices through the Model Context Protocol (MCP). Android devices connect over USB or Wi-Fi through ADB, while iPhone control uses Apple's developer services. A gateway applies approval rules, including requiring confirmation for payments and hiding banking and password applications by default; each run is recorded locally with screenshots. It is available for macOS, Windows, and Linux.

Mentioned in
1 video
Kind
AI
PNo. 7919
AIAINotes.us AI product

Plannotator

Open source · backnotprop/plannotator

Plannotator is an open-source, local browser-based review surface for AI coding agents and personal assistants. It connects through plugins, hooks, commands, or MCP, opening a review interface when an agent proposes a plan, produces a document or HTML artifact, or completes work. Users can annotate plans, specifications, Markdown, HTML, agent messages, local changes, and remote pull or merge requests; feedback is sent back to the agent as structured input, while plan approval allows the agent to proceed and denial returns revision feedback. Its code-review workflow captures Git diffs or fetches GitHub and GitLab changes, provides a diff viewer with line comments and suggestions, and can send an approval or feedback response to the agent session. Plannotator Inbox provides a local window for agent questions, files, decisions, and guided reviews, with replies returning as later agent turns. The application stores plans, diffs, annotations, drafts, history, and configuration locally by default and does not collect usage telemetry; network access is used for features such as URL fetching, remote repository data, configured AI providers, and optional sharing. It is distributed as a binary with integrations for agents including Claude Code, Codex, Copilot CLI, Gemini CLI, OpenCode, Kiro, Droid, Amp, DeepSeek Harness, and Pi, and is dual-licensed under Apache 2.0 or MIT.

TypeScript
Stars
★ 13,129
Forks
1,315
SNo. 8019
AIAINotes.us AI product

SuperMimic

Open source · HybridRobotics/supermimic

SuperMimic is open-source software from UC Berkeley's Hybrid Robotics group for humanoid whole-body motion tracking and control. It trains a single differentiable world model that captures shared robot dynamics, then improves a policy by propagating gradients through short imagined rollouts and feeding the results back into a physics simulator, rather than relying on model-free reinforcement learning on large GPU clusters. The repository provides generalized-coordinate and maximal-coordinate G1 task configurations, scripts for training the world model and policy and playing back trained models, and is built on mjlab. It is released under the Apache License 2.0.

Mentioned in
1 video
Kind
AI

Links mentioned

🔒 Full analysis locked

Unlock more videos and the full analysis

A credit unlocks one video's full analysis for good — the build steps, the tools and how each was used, the methods behind every use case. Pro opens the whole library instead, and raises how many videos you can analyse a day.

Unlock full analysis — free

Transcript

Searchable transcript of Top Open-Source GitHub Projects : open-code-review, LobeHub, hyperframes, devvm & graphicle #309 — ManuAGI - AutoGPT Tutorials (15:46). Search for a phrase, then click its timestamp to jump straight to that moment in the video.

Captions sourced from the original video on YouTube, published by ManuAGI - AutoGPT Tutorials. The video, its captions and all related intellectual property remain the property of their respective owners; AINotes claims no ownership. Provided for research, accessibility and search — see the Transcript Notice and Copyright Policy.

00:00 Every week, developers release new open-source projects on GitHub. And this is our weekly GitHub project update video covering the best of them. Top trending open- source GitHub projects. This week, you'll discover useful and trending developer tools from AI code review to agent coding workflows. Without wasting time, let's get started. >> Before we jump into today's project updates, here's a quick announcement for everyone.

00:23 We've launched a brand new YouTube channel called AI Agent Studio dedicated entirely to AI Agent projects, tutorials, and tools. So, if you're interested in staying up to date with the latest AI agent open source projects, learning how to build your own agents, or exploring cuttingedge agent frameworks, make sure to check it out. Subscribe now to get weekly videos, in-depth guides, and real time project breakdowns.

00:49 The link is right there in the description. Don't miss it. All right, let's get into today's video. >> Project number one, Open Code Review. AI code review CLI from Alibaba. Open Code Review is an open-source AI powered code review CLI from Alibaba. Its hybrid design pairs deterministic pipelines with an LLM agent. It reads your git diff, and the agent reads full files, searches the codebase, and inspects changes, then returns line level comments.

01:19 A multi- language rule set catches issues like null pointer bugs, thread safety, XSS and SQL injection, and it can auto apply fixes. It works with OpenAI and Anthropic with plugins for clawed code and codecs. Install it and review your code. Project number two, Planetator. Visually review and annotate your agents plans. Planetator is an open-source local review surface for AI coding agents.

01:44 When your agent writes a plan, it opens in your browser. So you mark it up visually, highlight lines, leave notes, request changes, then send the feedback back to the agent with one click. The same canvas reviews code diffs and PRs, markdown and HTML artifacts, and it opens changes from Git, Jujutsu, or Git Butler. It runs locally, saves approved plans to Obsidian, and works with claude code, codecs, and open code.

02:10 Install it and review before you build. Project number three, Claude Trading Skills, a toolkit of trading skills for Claude. Claude Trading Skills is an open-source collection of skills that turn Claude into a systematic research assistant for equity investors and traders. Each skill is a skills MD covering one job, market analysis, stock screening, technical charting, back testing a strategy, tracking an economic calendar, or reading 13F filings.

02:38 So, the agent follows a disciplined framework instead of guessing. It also journals your thesis from idea to close. It's US focused, MIT licensed, and works with Claude Code, Codeex, and Cursor. It's for research, not financial advice. Install it and systematize your analysis. Project number four, this ChromeDev Tools-MPP. Give your coding agent real ChromeDev Tools.

03:02 ChromeDev Tools MCP is Google's official MCP server that connects your AI coding agent to a live Chrome browser through the ChromeDev Tools protocol. Instead of guessing why a page breaks, the agent drives real ChromeDev clicking, filling forms, and navigating with Puppeteer, then reads the console, inspects network requests, takes screenshots, and records performance traces for actionable insights.

03:27 Written in TypeScript, it runs with one npx command, and plugs into Cloud Code, Cursor, Copilot, and Gemini. It's Apache licensed. Add it and let your agent debug the browser. Project number five, Autoontology. A semantic layer that grounds agent queries in your data. Autoontology is NVIDIA's open- source semantic layer that grounds AI agents in your enterprise data.

03:49 So, they return the right numbers instead of plausible looking wrong ones. It builds a graph of your database context where data lives and what each term means from your metadata, then guides agents to generate deterministic certified text to SQL queries with provenence. It ships a web app, a REST API, and an MCP server with agent skills for setup, ontology management, and querying.

04:13 Runs on Postgress via Docker, Apache licensed. Deploy it and ground your agents. Project number six, Nashi Lobhub, a self-hostable AI agent workspace for your whole team. LobHub is a free self-hostable AI agent workspace, formerly Lo Chat, that treats each agent as a unit of work you create, configure, and combine. Not just a chat window, it acts as your chief agent operator, hiring, scheduling, and reporting on a whole team of agents running around the clock.

04:42 It supports many model providers, a plug-in and skills marketplace, knowledge bases, and a polished design with a one-click free deploy of your private instance built in Typescript with a desktop app. deploy it and run your AI team. Project number seven, Hyperframes. Turn HTML into video built for agents. Hyperframes is Hen's open-source framework for making videos in code built for AI agents.

05:06 You write each frame as HTML and animate it with Gap. Then Hyperframes renders it to video. No After Effects, no timeline app because it's all markup. An agent can author and revise every shot and bundled skills teach it the engine, timing, and idioms. It covers launch films, product demos, and explainers and plugs into claude code and cursor. It's MIT licensed.

05:32 Install it and generate video from HTML. Project number eight, Knowledge Work Plugins. Anthropics plug-in suite for Claude knowledgework. Knowledgework Plugins is Anthropic's open-source collection of plugins for knowledge workers using Claude, especially in Claude co-work and Claude code. Each plug-in bundles skills, commands, and settings for a role or function areas like productivity, sales, marketing, finance, legal, and data analysis, so Claude follows a consistent expert workflow.

06:02 They're organized by team, installable individually, and built to be customized. Browse the repo and add the plugins your work needs. Project number nine, modern web guidance. Teach your agent to build with modern web platform features. Modern web guidance is Google Chrome's open-source skill that keeps AI coding agents building with current web platform features instead of outdated workarounds.

06:25 It packs more than 100 expert verified guides covering performance, accessibility, responsive UI, forms and security, pass keys and browser capabilities and checks them against baseline support data. So the agent reaches for native HTML and CSS with safe fallbacks. A CLI searches and installs it and it plugs into Cloud Code, Codeex, Gemini, and Copilot backed by the Chrome and Edge teams.

06:50 It's Apache licensed. Install it and build modern project number 10. Exact. Build one native app across platforms for agents. Exact is Expo's experimental cross-platform application runtime designed so a coding agent can build and test one app everywhere. You describe the interface in contract, a small declarative language and write logic in TypeScript or Rust and Exact renders it natively.

07:16 The real DOM on web app kit on Mac OS UAC on iOS and GPU rendering on Linux from a tiny runtime. It's built for agents, constrained languages, visual inspection, portable tests, and a deterministic clock that kills flakes. It's pre 1.0 under active development. Clone it and build everywhere. Project number 11. Vine dynine do super mimic. Teach humanoids whole body control with a world model.

07:42 Supermic is the open- source code from UC Berkeley's hybrid robotics group for scaling humanoid whole body control with one differentiable world model. Rather than running model free RL on huge GPU clusters, it learns a single world model that captures the robot dynamics every skill shares, then improves the policy with gradients back propagated through short imagined rollouts, looping back to a physics simulator to collect data.

08:07 One humanoid learns to walk, jump, and get up while keeping balance. Trainable on a single GPU. Clone it and train a humanoid. Project number 12. Henge. Run Mac OS on your iPhone and iPad. Henge is an open-source project that runs Mac OS on an iPhone or iPad. Since iOS gives apps no hypervisor, it ships its own emulator. A custom JIT runs the Mac's ARM 64 code with its own MMU and virtual devices.

08:36 No QMU, no virtualization framework. It installs Mac OS straight from Apple's restore image with no Mac. Draws the Mac's GPU through Metal. Turns the screen into a trackpad and shares clipboard and files. It needs a recent chip, lots of storage, and sideloading. It's GPL licensed and early. Sideload it and run Mac OS in your pocket. Project number 13, My Go Music, a native Proden extensible desktop music player.

09:03 Migo Music is an open-source native desktop music player written in Go, no Web Zoo that you extend with plugins. It plays your local files out of the box with tags, covers, and LRC lyrics, playlists, liked songs, search, and a now playing view with lyrics styles, and a cover wall. Plugins add other sources, accounts, cataloges, podcasts, and stations.

09:28 It brings a Mac OS notch player, media keys, loudness normalization, and ondevice podcast captions. Plugins are JavaScript or TypeScript. It's AGPL licensed. Build it and play your music. Project number 14, Devie FVM, a dedicated isolated VM for every project. DevM is an open-source tool that gives each project folder its own isolated virtual machine, one VM for every project.

09:55 Run devm in a folder and it detects the tool chains your project needs. Builds a sandbox VM with them. Mounts only that folder and drops you into a shell. It's made for coding agents. Run clawed code, codecs or open code inside the VM so they never touch your host and your tokens stay on your machine. VMs clone a shared base so new ones start in seconds.

10:18 Built and go on Lima MIT licensed. Install it and sandbox your project. Project number 15, CUI. Native UI components for C across every platform. CUI is an open-source native user interface library for C, a toolkit of around 150 components, including buttons, form fields, dialogues, tables, charts, planning boards, and AI chat elements. Each component is a C function that draws itself, handles input, and reports what happened while your own variables hold the state.

10:51 So the interface never needs separate state syncing. It runs natively on Windows, Mac OS and Linux from one codebase using Win32, Coco and X11 backends or SDL2 and GV elsewhere with no dependencies and a footprint near 240 KB. It renders through OpenGL 3.3 or a software fallback and builds with CMake. Grab it and build your native app. Project number 16, Pinky.

11:16 Let AI agents operate your real phone safely. Pinky is an open-source desktop app that lets AI agents like Claude Code, Codeex, or Cursor operate a real iPhone or Android phone through MCP with a command line helper that agents launch as an MCP server. Android phones connect over USB or Wi-Fi through ADB, and iPhones connect through Apple's developer services.

11:41 A gateway enforces approval rules, so an agent can approve its own actions. Payments and transfers wait for your click, and banking, wallet, and password manager apps are hidden by default. Each run is recorded with step descriptions, screenshots, and a screen recording. All stored on your computer. Available for Mac OS, Windows, and Linux. Install it and put your agents to work on your phone.

12:04 Project number 17, Chrome Capture. Record a Chrome Tab at high frame rates. Chrome Capture is an open-source CLI and npm library for recording a Chrome tabs video and audio straight to an AV1 MP4 at high frame rates and resolutions. Older recorders in Playright and Puppeteer rely on a screencast API that ships JPEG frames over the DevTools protocol into FFmpeg, which caps the frame rate and makes stopping slow.

12:33 Chrome capture instead uses Chrome's newer screen recording API, sending frames from the compositor to an encoder with no FFmpeg needed and near instant stops. It plugs into Playright and Puppeteer page objects, raw CDP sessions, and a test fixture that attaches recordings to reports or runs from the command line. Install it and record your browser sessions.

12:56 Project number 18, Phone Use. View and control an unmodified iPhone from Linux. Phone Use is an open- source technology demo that lets a Linux PC view and control an unmodified iPhone with a command line interface, an MCP server, a browser mirror, and a ready-made skill for clawed code agents. It needs no jailbreak, developer account, cable, or app on the phone.

13:18 Only the built-in screen mirroring and assisted touch features. The phone screen arrives over AirPlay as a desktop window and the PC acts as a Bluetooth mouse and keyboard. So clicks become taps and drags become swipes. Agents take screenshots, read the screen as text with positions, then tap, swipe, type, or open apps. It builds on PlayPlay, Blue Z, Greamer, GTK4, and Python.

13:42 Grab it and automate your iPhone. Project number 19, Graphical, a code map and memory for AI coding agents. Graphical is an open- source tool that gives AI coding agents a map of your code and a memory of your project. Shipping as one Rust program that runs as an MCP server alongside a command line interface and a browser app. Agents often reread large parts of a repository to find relevant code and lack your project's docs, decisions, and past sessions.

14:12 Graphical syncs your folder respecting doc get ignore and skipping secrets and builds a code graph of files, functions, calls and imports using tree sitter. No model required. It indexes docs and reviewed memories, captures past claude code and codeex sessions and returns cited excerpts with file and line numbers within a token budget while git hooks keep the index current.

14:35 It connects with Claude Code, Cursor, Codeex, VS Code, Claude Desktop, and any MCP client and runs locally with no account or telemetry. Install it and give your agents real context. Project number 20, Ding. A single Toast notification that morphs in place. Ding is an open-source toast notification library for web apps written in Typescript. Most toast libraries stack multiple notifications on screen.

15:00 Ding instead morphs one toast in place when a new message arrives and its animations run on a spring solver, so interrupting one keeps its momentum. A core package handles springs, timers, state, sound, and the plain DOM renderer, while framework packages handle only rendering and styles are injected from JavaScript. So there's no stylesheet to import.

15:21 It works with React 18 plus, spelt 5, and plain DOM. You install one framework package which pulls in the core. It supports light, dark, and system themes, six screen positions, and right to left layouts with no runtime dependencies and built-in accessibility features. Install it and add notifications to your app. Thanks for watching. See you in the next update.