Agent Desktop is a Rust-based CLI and C library for letting AI agents inspect and operate desktop applications through operating-system accessibility trees rather than pixel matching. It returns JSON snapshots containing stable, snapshot-scoped references for actionable elements; agents act on those references and take another snapshot to verify the result, while stale or ambiguous references produce structured errors. It supports macOS, Windows, and Linux, with commands for application and window management, keyboard and mouse input, clipboard access, notifications, screenshots, sessions, and batch operations. It can be installed from npm with platform-native binaries or used through release archives and its FFI library. macOS requires Accessibility permission, Linux requires a graphical session with at-spi2-core, and the project is licensed under Apache-2.0.
Chimera is a browser-based physical-modeling voice synthesizer. It generates sound from a simulated airway—including lungs, vocal folds or a syrinx, branched vocal tract, lips, and nose—rather than from samples. Its Rust engine is compiled to WebAssembly for an AudioWorklet and can also run natively through the command-line interface. The browser interface supports live articulation by dragging the tongue, lips, and jaw in a cutaway head view, as well as vowel pads, encoders, keyboard input, and Web MIDI. It turns English text or IPA into vocal gestures and prosody, provides a piano roll for singing, and uses articulatory gestures for beatboxing. Additional modes model songbird, seal, alligator, and elephant voices; male, female, and child bodies can be blended. The CLI supports offline rendering and diagnostics for speech, songs, notes, beatboxing, creatures, and vowels. Chimera is distributed under the MIT License. The repository also includes tools and documentation for acoustic modeling, real-time engine behavior, spectrograms, and comparisons with recorded singers and speakers.
flawless-ui-ux is a UI/UX skill and command-line tool for coding agents that build, redesign, refine, and review web interfaces in an existing project. The skill connects product context with composition, typography, color, density, imagery, and interaction, while the CLI collects project context, inspects selected source files or a running interface, and compares later results after corrections. Browser inspection uses Playwright and selected axe accessibility rules to check items such as contrast, accessible names, computed font sizes, horizontal overflow, broken images, uncaught JavaScript errors, repeated phrases, and arrow or icon cues. It can save inspection artifacts and screenshots, then classify findings as persistent, new, or no longer observed during verification; its samples do not establish full WCAG conformance or complete keyboard and business-flow correctness. The package can be installed with npm, pnpm, Bun, a skills CLI, or its Bash installer, requires Node 22.12 or later for the package, supports project or global installation for multiple coding-agent hosts, and is distributed under the MIT license.
An open-source Claude skill and small TypeScript library for animated line-art SVG components. It draws ink-on-paper figures and scenes live in SVG, including walking, riding, reactions, day-to-night palettes, pointer-driven interactions, and elements such as wind, reflections, windows, and lanterns. The repository includes a workflow and style rules in SKILL.md, four dependency-free TypeScript modules for drawing primitives, gait solving, springs, pendulums and wind, and time-of-day palettes, plus walker and village examples. Its gait solver uses planted-foot motion, a minimum-jerk swing path, heel lift, and hip-height adjustment; scripts use headless Chrome to capture deterministic frames for video and to measure or compare reference clips. The examples require Node 22.18 or newer, and the project is licensed under MIT.
MousePods is an open-source Linux tool that turns motion sensors in AirPods 3 and AirPods Pro into a head-tracking mouse or simulator controller without an Apple device. It connects over Bluetooth through BlueZ, starts the appropriate sensor protocol—Apple RemoteHID for AirPods 3 or SensorDataWX for AirPods Pro—and sends the resulting motion data through a virtual uinput device for Wayland and X11. Its motion engine derives yaw, pitch, and roll from gyroscope and gravity data, re-estimates gyro bias while stationary, and applies One-Euro smoothing; cursor modes include head or bud movement, calibrated screen pointing, and experimental inertial gliding. The local desktop app provides battery and orientation displays, live sensor data, calibration, settings, shortcuts, and a sensor explorer, with a Flatpak distribution available. It also supports head tracking in Proton games through opentrack and includes an experimental Chrome extension for controlling 360-degree video views. The source code is licensed under Apache-2.0; protocol documentation, findings, and captures use CC BY 4.0.
Native Shell is an open-source, Rust-based cross-platform UI framework for iOS, Android, macOS, Windows, Linux, terminals, GPUI apps and the web. Applications can be written in Rust, TSX with Tailwind classes, or a combination of both; NSH shares components, state, APIs and a reactive layout engine while rendering through platform-specific backends such as UIKit, Android views, AppKit, Win32, GTK, ratatui or HTML. It supports standalone applications, static websites and embedding inside existing native applications. NSH can deliver TSX screens at runtime as signed packages, including plugins, feature modules and AI-generated interfaces. Host applications configure the capabilities available to these modules, which run in a sandbox. The project also provides hot updates for TSX, generated TypeScript declarations, headless tracing and an `nsh` CLI for creating, running, building and packaging applications across its supported targets. The repository is licensed under Apache License 2.0. Documentation, guides and live demos are provided at nsh.dev.
OpenMausBot is an open-source, local-first chat application for managing a team of AI bots as separate contacts. Each bot can have its own personality, model, conversation thread, computer, and connected applications, using agents launched through locally installed Claude, Codex, or Grok command-line tools and existing user logins. A local harness server on 127.0.0.1 manages agent processes, transcripts, keys, and events in the user's local OpenMausBot data directory. Bots can work through a cloud Linux desktop, an isolated local virtual machine, or—where supported—the host computer. Shell commands, file edits, and questions are routed through a permission broker and shown as approval or response cards. The application includes model selection, bot and conversation management, browser-accessible desktop takeover, and connected applications through Composio, including services such as Gmail, Slack, GitHub, Notion, and Linear. The repository provides desktop distributions for macOS, Windows, and Ubuntu, with Ubuntu Wayland host control disabled according to the README while a stated issue is resolved.
PulsoIcon is an open-source animated icon library by CityPulso, providing 9,955 icons for React, Vue, Svelte, Solid, Preact, Angular, Astro, Jetpack Compose, and SwiftUI. It is designed as a drop-in replacement for lucide-react in React, using the same component names and props, while also incorporating shapes from Lucide, Tabler Icons, Iconoir, Hugeicons, Phosphor, and Simple Icons. Each icon has its own hand-written motion rather than a shared animation effect. Icons play when the surrounding button, link, label, tab, menu item, or another configured trigger is hovered or focused; they can also be controlled programmatically. The packages are tree-shakable, with each icon in its own side-effect-free module, and support reduced-motion preferences, static rendering, solid variants, aliases, and consistent icon data across platforms. The project provides framework-specific packages, a shared core package, generated platform implementations, documentation, galleries, tests, and build scripts. It is distributed under the ISC license, with license-notice requirements for shipped applications and derived icon sets.
Rune is a native UI framework written in Rust that renders interfaces on the GPU. It includes its own renderer, text engine, layout system, and spring-based animations, and targets macOS, Windows, Linux, and the web from one codebase. Applications can run headlessly, allowing a coding agent or other scripted process to render screens to PNG, inspect the element tree, and drive controls with scripted input. The project is pre-release, uses an MIT license, and is developed by Anomaly.
Sakutto is a free, MIT-licensed Chrome and Firefox browser extension for translating web pages, selections, PDFs, video captions, editable text, and typed text on-device. Its default engine is the PLaMo 3 Translate model running through a pinned wllama build with WebGPU or CPU; it supports PLaMo's 53 languages, bilingual page views, glossaries, Easy Japanese, and keyboard shortcuts. Chrome can use its built-in translator for 11 additional target languages, while Firefox uses PLaMo only. The extension requires no account, advertising, analytics, or telemetry, keeps translation data on the device by default, and downloads the model once for offline use. It is an unofficial project and is not affiliated with Preferred Networks.
satmux is a pure-Python TCP multiplexer and HTTP/TLS relay for high-latency geostationary satellite inflight Wi-Fi. A local client maintains a pool of persistent, non-443 TCP connections to a Linux VM near the satellite ground station, using BBR, 1280-byte MSS, and zlib compression; streams are scheduled across the carriers to avoid repeated handshake delays and cabin :443 proxy failures. It exposes local SOCKS5 and HTTP CONNECT proxies, as well as a 1-RTT encrypted ground-fetch path that performs DNS, verified TLS 1.3, redirects, and response compression near the origin network. The repository also includes routing for Git and the gh CLI through a local Unix socket and macOS setup scripts. It uses only the Python 3 standard library, requires an authenticated cabin Wi-Fi session and a separately rented ground VM with an open TCP port, and is licensed under MIT.
Sophon-Realtime is an open-source family of full-duplex voice models from Adventists AI. It continuously listens while a person speaks, uses a speech model to decide whether the speaker has finished, is back-channeling, or is interrupting, and drafts a reply from pauses before releasing audio. Its reply can include expressive sound events such as laughter, sighs, breaths, coughs, and emphasis; barge-in handling discards an interrupted draft and continues with the new question. The family includes Tag, Flash, and E2E builds. Tag and Flash generate sound-tagged replies that an expressive speech model performs, while E2E generates continuous voice vectors directly with a compact talker and decoder. The models support speech-to-speech interaction, tool calling, OpenAI Realtime-compatible interfaces, zero-shot voice cloning from a short clip, and a Live mode that delegates lookups and actions to a user's backend. They can run locally with vLLM-Omni on CUDA or with MLX on Apple silicon, or be accessed through Adventists AI's hosted API. Code and weights are distributed under Apache-2.0.
Swarm Rebuild is a Chrome extension that replaces a web page with a blank canvas and reconstructs it live using a swarm of multiplayer cursors, including drawn boxes, typed text, pen-tool lines, ruler guides, media, and duplicated elements. It is a tribute to Joel Califa's Swarm website. The extension snapshots the live DOM, including open shadow roots, into a blueprint of styled boxes, text layers, borders, media, repeated elements, and alignment guides. It then overlays the page with an isolated layer and uses task-specific cursor scripts to rebuild the snapshot as the user scrolls; duplicates wait for their master elements, and the swarm uses orbiting, alignment, and mouse-avoiding flocking effects. Synthesized sound effects are generated with the Web Audio API rather than audio files. It can be activated from the toolbar or with Alt+Shift+S, supports replay, live-page restoration, variable build speeds, and an optional sound mode. The repository also includes a self-contained test page and instructions for loading the extension unpacked. It is dependency-free, distributed as a Manifest V3 extension, and has no license yet; the project notes that parts of the original Swarm site were adapted and reuse should be checked with its creator.
T3 Code is an open-source agent-harness control surface for coding agents running on a developer's computer. It provides iOS and Android mobile apps, a web app, and an Electron-based desktop app for controlling locally configured Codex, Claude Code, Cursor, Grok Build, and OpenCode agents while using the user's existing provider subscriptions. The project can be launched with `npx t3@latest`, which starts its backend and local web interface on the machine. It also distributes desktop builds for Windows, macOS, and Arch Linux, and supports remote access from a phone or another machine. The repository describes the project as early-stage and warns that bugs are expected.
Uptime Kuma is an open-source, self-hosted monitoring tool for checking website and API availability, TCP services, keyword and JSON responses, WebSockets, ping targets, DNS records, Steam game servers, Docker containers, and other endpoints. It can also monitor TLS certificate information and, as described in the video, databases and firewalls. It provides a web interface with uptime charts, certificate details, proxy and two-factor authentication support, and multiple status pages that can be mapped to specific domains. When checks fail, it sends notifications through Telegram, Discord, Gotify, Slack, Pushover, email via SMTP, and more than 90 other notification services. The project can be deployed with Docker Compose or a Docker image, or run directly on supported Linux and Windows systems with Node.js, Git, and optionally PM2. The repository documents monitoring intervals of 20 seconds and warns that network file systems such as NFS are not supported for its data directory.
Veyra is an open-source Windows video player and capture-enhancement tool for local files, images, capture cards, PS5, PC, Xbox and screen sources. It builds a real-time GPU processing pipeline with layered neural denoising, super resolution, HDR and colour effects, anti-aliasing, and optional frame generation; it supports NVIDIA and AMD paths plus DLSS, XeSS and FSR options. The application uses Qt Quick/QML for its interface and a native Direct3D 12 video renderer. Its Professional mode provides either an ordered List pipeline or an executable Node graph with connected processing stages; multiple denoising layers have independent settings, while super resolution is a shared stage and frame generation is applied at the end of the live path. It also provides capture and streaming inputs, screenshots, and an offline export queue for MP4/MKV files with trimming and selected audio or subtitle tracks. Veyra is distributed as Windows 11 x64 portable packages for NVIDIA or AMD GPUs and requires DirectX 12. The original code is licensed under GPL-3.0; the combined Chiaki remote-play component has separate AGPL-3.0 licensing with an OpenSSL exception. The repository notes that some runtimes and models are proprietary and that the project is not a complete official DLSS 5 integration.
VoiceStudio is an open-source, fully local desktop application for voice cloning and design, text-to-speech, transcription, dictation, video dubbing, and long-form audio production. It runs on macOS Apple Silicon, Windows, Linux, or Docker and keeps voices, projects, settings, and generated files on the machine by default. The application combines a Tauri desktop shell, a React and Vite interface, and a local FastAPI backend with registries for multiple TTS and ASR engines. Its workflows can transcribe and translate video, preserve speaker assignments, synthesize replacement speech, render audiobook chapters, separate vocals, diarize speakers, process batch jobs, and route work to CPU, CUDA, Apple Silicon MPS/MLX, ROCm, or optional remote workers. It also exposes local REST, SSE, WebSocket, OpenAI-compatible audio, and MCP interfaces, including speech synthesis and transcription endpoints. VoiceStudio is licensed under AGPL-3.0; downloaded models and tokenizers retain their own licenses. The local workflow requires no account, API key, subscription, or usage meter, although remote workers, configured external ASR endpoints, analytics, and other network-backed functions are opt-in. The repository describes the software as an active beta and distributes packaged desktop releases as well as source code.
vork is a headless 3D and VFX engine for AI agents, developed by Voyager. Agents define scenes as code, including geometry, shaders, materials, cameras, physics, and compositing, then render images or video, inspect the results, and revise the scene without an interactive editor. The engine is written in Rust and Slang and renders on the GPU through Vulkan or Metal, combining raster visibility with hardware ray-traced lighting; an optional path tracer is also available. Public program hooks let scene authors customize camera rays, material response, light propagation, film response, and GPU simulations such as fluids or agent-based motion. Scenes can be authored as JSON or through Python and TypeScript SDKs, and the command-line tool provides project initialization, rendering, help, profiles, and output of images, video, and control passes. vork is source-available and free during its research preview, with binary packages for Linux, Windows, and Apple Silicon macOS. Rendering requires a compatible NVIDIA ray-tracing GPU or Apple Silicon GPU; the preview has limitations including unsupported participating media, possible failures on large or transparent scenes, and potentially lengthy compilation when new GPU programs are first used.
Searchable transcript of Top Open-Source GitHub Projects : uptime-kuma, VoiceStudio, buzz, nsh, satmux & sakutto #310 — ManuAGI - AutoGPT Tutorials (14:18). 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 coding agents to local ondevice apps. Without wasting time, let's get started. >> Before we jump into today's project updates, here's a quick announcement for everyone.
00:24 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.
00:50 The link is right there in the description. Don't miss it. All right, let's get into today's video. >> Project number one, Uptimekuma. A fancy self-hosted Uptime monitoring tool. Uptimekuma is an open-source self-hosted monitoring tool that watches your services and alerts you the moment they go down. It checks HTTPS, TCP, ping, DNS records, websocket, Docker containers, and certificate info at 20-se secondond intervals, then pushes alerts through Telegram, Discord, Slack, email, and over 90 other services.
01:24 It ships status pages you can map to your own domains, a reactive view 3 UI with ping charts, 2FA, and multi- language support. Deploy it with one Docker command, and monitor everything. Project number two, Mens Buzz. A workspace where humans and agents build together. Buzz is an open-source self-hostable workspace from Block where humans and AI agents share channels and build together on a relay you own.
01:51 It combines chat, code hosting, code review, CI results, workflows, and project search in one system where every message, reaction, review approval, and git event is assigned event in a single log. Agents get their own keys, channel memberships, and audit trail. So you treat them like team members. Built in Rust on the Nost protocol with a Tory desktop app.
02:13 Self-host it and build together. Project number three, Vera NR video, AI upscaling and frame generation for Windows video. Very is a free open-source Windows video player and capture card enhancement tool that applies AI based upscaling, image enhancement, and frame generation to video files, images, and live capture card input, and can stream from a PS5 over your local network.
02:36 It runs DLSS and RTX video super resolution up to 8K, plus DLSS 2x4x frame generation. PS5 remote play, HDR10 export, and sideby-side comparison are builtin. Built in C++ 20 on Direct 3D12. It needs a compatible RTX GPU. Grab it and enhance your video. Project number four, Mance Open Mossbot. Your own team of AI bots in a chat app. Open Mosbot is an open-source alternative to Grockbot that gives you your own team of AI bots in a chat app.
03:09 Each bot in the sidebar is a real agent claude codeex or grock running locally on your machine with its own personality model cloud computer and connected apps. It brings perbot model switching mid conversation inline approval cards for shell commands and edits and connected apps like Gmail, Slack, GitHub and notion built-in TypeScript and React on Electron.
03:34 Install it and run your bot team. Project number five, Nichch RS need neml tornep give AI agents computer use on desktop apps. Agent desert is an open-source command line tool and C library that gives AI agents computer use on desktop apps by reading each app's OS accessibility tree instead of screenshots or pixel matching. The agent reads the UI as JSON acts on elements by reference then rereads the UI to verify the tool itself contains no model.
04:02 It runs headless by default, exposes every command as an MCP server, and ships around 57 commands for observation, input, and window control, built-in Rust for Mac OS, Windows, and Linux. Install it and automate your desktop. Project number six, D3 Code, a control surface for your AI coding agents. T3 Code is an open-source agent harness control surface for running AI coding agents on your own machine built by ping.gg GG to give a better development experience with agents than existing tools offered.
04:36 It works with cloud code, codec, cursor, gro, open code, and Google anti-gravity. A command line launcher starts a local server and opens the web app with iOS and Android apps, remote access from your phone, permission modes, and per project settings. Built-in TypeScript with an Electron desktop app. Install it and command your agents. Project number seven, Voice Studio.
04:58 A fully local, open-source 11 Labs alternative. Voice Studio is an open-source, fully local AI Voice Studio. An 11 Labs alternative that runs on your own hardware with no account for the core app. It handles voice cloning from a reference recording, voice design from a text description, video dubbing, dictation, local transcription, and audiobook narration with local models managed through a model manager and support for 646 languages.
05:26 It ships a local API, MCP support, a CLI, and watch folders. Built as an Electron app with a Python backend. Install it and generate speech locally. Project number eight, Claude Plaus Official. Anthropic's official plug-in directory for Claude Code. Claude Plus Official is Anthropic's official managed directory of highquality plugins for Claude Code.
05:51 It's split into internal plugins developed and maintained by Anthropic and external plugins from partners in the community, which must meet quality and security standards for approval. You install any plug-in with a single /comand or browse them through discover and each can bundle/comands, agent definitions, skills, and MCP server configuration. The readme reminds you to trust a plug-in before installing it.
06:13 Browse the repo and add the plugins your workflow needs. Project number nine, line art animation. A claude skill for animated line art scenes. Line art animation is an open-source Claude skill and small TypeScript library for animated line art components. It draws ink on paper figures that walk, ride, and react in scenes that move from day to night, rendered live in SVG, so they stay sharp at any size and respond to the pointer.
06:39 It ships modules for a walk solver, springs and wind, and a time of day pallet, plus example scenes, and deterministic frame capture for video. Video rendering runs through headless Chrome with FFmpeg. The library modules have no dependencies. Install it and animate line art. Project number 10, Rune, a native UI framework. Rune is an open-source project from the Anomaly Co.
07:04 account described as a native UI framework. Its repository route contains cargo files which points to a Rust project. Though the page doesn't confirm the stack, it's MIT licensed and the repo includes crates, examples, a demo app, theming, and a developer tool along with instruction files for AI coding agents. Clone it and build a native UI. Project number 11, Nishn AI generated UI native in your app.
07:30 Native shell is an open-source framework from Longbridge for building UI with TSX and Tailwind that runs with each platform's native widgets. It lets apps load UI modules at runtime as plugins written by an AI model or a team or package them as standalone apps. Each view runs sandboxed in its own QuickJS VM with signed ED2519keyed packages. The core is Rust spanning iOS, Android, Mac OS, Windows, Linux, and web.
08:00 A single CLI handles bundling, hot reload, and packaging. Install it and render native UI. Project number 12. Dish Swarm Cursors. Rebuild any web page with a swarm of animated cursors. Swarm Rebuild is an open- source Chrome extension that wipes a web page to a blank canvas, then rebuilds it with a swarm of animated cursors. A visual rebuild, not a functional clone.
08:23 It's a tribute to Joel Khalifa's Swarm, contributing a way to snapshot any live DOM into a blueprint the Swarm can rebuild. Cursors drag Rex, type text, draw pen tool lines, and snap to ruler guides with live synthesized sound panned to each cursor. It's dependency-free JavaScript. Load it and watch pages rebuild themselves. Project number 13, Vorc, a headless 3D and VFX engine for agents.
08:50 Vorc is a headless 3D and VFX rendering engine designed for AI agents. In research preview, agents write scenes as code, render them on a GPU, inspect the output, and revise. There's no graphical editor. Scenes are defined in JSON with optional Python and TypeScript SDKs. And render profiles span fast previews to final path tracing. It outputs stills, video, and control passes with a GPU fluid library and glTF imports.
09:18 Built in rust and slang on Vulcan and metal. It's source available and free during the preview. Clone it and render in code. Project number 14, a TCP multiplexer for high latency inflight Wi-Fi inflight is stationary inflight Wi-Fi like in Marsat GX, Vasat and Panasonic. It speeds up slow connections mainly by removing satellite round trips and avoiding cabin proxies.
09:52 It keeps a pool of 12 pre-warm TCP connections to a Linux VM near the ground station with local socks 5 and HTTP proxies and a groundside fetch service that handles DNS, TLS, and compression. It even wires into Git and GH. It's pure Python 3 with no dependencies. Install it and speed up your link. Project number 15, mouse pods. Turn your AirPods into a mouse on Linux.
10:17 Mouse Pods is an open-source Linux tool that uses AirPods 3 and AirPods Pro motion sensors as a headtracking or air mouse input device with no Apple device needed. It also publishes what it says is the first public documentation of Apple's remote ATI protocol. It computes yaw, pitch, and roll, drives a virtual U input device at 120 hertz on Whan and X11, and offers mouse, pointer, and an experimental glide mode with stem press and dwell clicks.
10:47 A local web app shows battery, live 3D orientation, and calibration. Built-in Python on Blue Z. Install it and point with your head. Project number 16, Pimemer Sifken Synth, a browser voice synth built on physical modeling. CHIM is an open-source browser-based voice synthesizer that uses physical modeling rather than samples. The sound comes from a simulated airway, including lungs, vocal folds, a branched vocal tract, lips, and nose.
11:17 You can play it by dragging the tongue, lips, and jaw in a live head cutaway. Speak English text or IPA. Sing from a piano roll and beatbox with drum pads. Presets even model a song bird, seal, and elephant. The engine is Rust compiled to web assembly. Open it and synthesize a voice. Project number 17. Pulso icon package. Animated React icons. A drop in Lucid replacement.
11:42 Pulso icon is an open-source library of animated icons. A drop-in replacement for Lucid React built on motion. Each icon animates when users hover or focus on the control it belongs to with its own motion instead of a shared effect. It ships about 9,955 icons and works across React, Vue, Spelt, Solid, Angular, Astro, Jetack, Compose, and Swift UI. Animations respect reduced motion, and each icon is its own tree shakable module built-in TypeScript.
12:13 Install it and animate your icons. Project number 18, Flawless UIUX, a UIUX skill for coding agents. Flawless UIUX is an open-source UIUIUX skill for coding agents to build distinctive usable interfaces. A skill guides the design work while a local CLI collects evidence and checks what changed after corrections. It inspects a running app at a URL across default viewports, saves a baseline, and checks accessibility, text size, horizontal overflow, broken images, and uncaught JavaScript errors.
12:48 Inspection never edits your sources or submits forms. It plugs into codecs and open code and runs on a bun tool chain. Install it and refine your UI. Project number 19. Sophon Realtime. Open full duplex voice models that hear your tone. Son Realtime is an open- source family of full duplex voice models from Adventist AI that hear your tone and speak with one.
13:11 They listen while you talk, pick up how something is said, and reply with expressive speech like laughs, size, and breaths. Three builds share 132B brain, reaching first audio in around.3 seconds with bargin tool calling and a transcript on every reply. Built in Python on CUDA, served with VLLM, Apache 2.0 and free for commercial use. Deploy it and build voice agents.
13:38 Project number 20, Sakudo ondevice webpage translation in your browser. Sakudo is a free open-source browser extension for Chrome and Firefox that translates web pages entirely on your device. Running the Playo3 translate model through W Llama, no account, ads, telemetry, or translation server with the only network request being the model download. It translates whole pages, selections, and type text with bilingual view, shortcuts, and right-click menus.
14:06 Playo supports 53 languages in either direction. Built-in TypeScript with spelt and webgu. Install it and translate privately. Thanks for watching.