Top Open-Source GitHub Projects : Flutter, VoiceStudio, Splash, KCC, upm & disktree

Author: ManuAGI - AutoGPT Tutorials · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: Top Open-Source GitHub Projects : Flutter, VoiceStudio, Splash, KCC, upm & disktree · Channel: ManuAGI - AutoGPT Tutorials · 13:31 · Watch on YouTube

Topics: AI agents, Software engineering with AI, AI content generation, Local and open-source models, Open source and software ecosystems

Summary

The video presents 20 open-source projects, ranging from AI agent tools and local inference software to app development frameworks and utilities. It highlights projects such as OpenShell for sandboxing AI agents, Voice Studio for local voice tools, Flutter for cross-platform apps, and Splash for local inference on Apple Silicon. Other projects support parallel AI coding workflows, e-reader comic conversion, software design, disk cleanup, and package management. The video closes with tools for explaining UX decisions and making image-based decisions with a locally run model.

Key points

OpenShell provides a policy-governed sandbox runtime for AI agents, with credentials restricted to approved requests.
Firebase Apple SDK offers Firebase services for Apple platforms, while Flutter supports app development across mobile, web, and desktop from one codebase.
Impeccable adds design commands and deterministic checks to AI coding agents; Voice Studio provides local voice cloning, dubbing, transcription, and audiobook creation.
Octtop is a self-hosted multi-user AI assistant, and Work Trunk manages Git worktrees for parallel AI agents.
Splash runs local inference on Apple Silicon and supports coding agents and compatible apps.
KCC converts comics and manga for e-readers, while Whiteboard lets people and AI agents collaborate on software architecture.
Disctree visualizes disk usage to help users find and remove large files and folders.
The list also includes tools for code-generated art, npm package management, UX explanations, and image-based decision-making.

Chapters

Start	Chapter	Summary

Quotes

You declare what each agent can touch and open shell enforces it. — 01:02

Everything runs on your machine with no accounts, API keys, or cloud. — 03:09

One codebase ships to iOS, Android, web, Windows, Mac OS, and Linux. — 04:09

Every beat grows out of the one before with no cuts. — 07:16

Concepts and entities

OpenShell, Firebase Apple SDK, Impeccable, Voice Studio, Flutter, Worktrunk, Splash, KCC, OpenAI MCP Extensions, Whiteboard, Disktree, Anodoodle, Naive N0.5 NMLE, Flutter

Full transcript (original language)

00:00 Every week, developers [music] release new open-source projects on GitHub. And this is our weekly GitHub project update video covering the best of them. Top trending [music] open source GitHub projects. This week, you’ll discover useful and trending developer tools from local AI [music] engines to cross-platform SDKs. Without wasting time, let’s get started. >> Before we jump into today’s project updates, here’s a quick announcement for everyone. We’ve launched a brand new YouTube channel called AI Agent Studio dedicated entirely to AI Agent projects,

00:32 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 realtime project breakdowns. The link is right there in the description. Don’t miss it. All right, let’s get into today’s video. >> Project number one, OpenShell secure policy governed sandbox runtime for AI

01:02 agents. OpenShell is Nvidia’s open-source secure by design runtime for autonomous AI agents. Agents like clawed code, codeex, and open claw run unmodified inside kernel level sandboxes governed by declarative YAML policy. You declare what each agent can touch and open shell enforces it. Agents never see credentials. It injects them only into requests bound for approved endpoints, blocking exfiltration. Policy changes are formally verified, so risky access waits for review. It runs on prem in

01:34 Docker or Kubernetes with SDKs. Deploy it and sandbox your agents. Project number two, Firebase Apple SDK. The open-source Firebase SDK for Apple platforms. This is the open-source Firebase SDK for Apple app development. the source for its Apple libraries. Firebase is Google’s app platforms and you add what you need. Authentication for signin, cloud fire store and real-time database, cloud storage, cloud messaging for push notifications, crash litics for crash reporting, remote

02:06 config, app check and firebase AI logic for Gemini. It targets iOS, iPad OS, Mac OS, tvOS, and watch OS and installs through Swift Package Manager, Coco Pods, or Carthage. Add it and wire Firebase into your app. Project number three, Impeccable, design skill that sharpens AI coding agents. Impeccable is an open-source design skill that makes AI coding agents better at front-end design. It installs into Clawude Code, Cursor, Codeex, Gemini, and more, adding

02:38 23 commands. Audit to find issues, polish for cleanup, distill to remove complexity, critique for a full review, each able to focus on a specific area. It runs 59 deterministic checks that catch the usual AI design tells like inter everywhere and purple to blue gradients and iterates live in the browser. Install it and sharpen your interfaces. Project number four, Voice Studio. Fully local, open-source alternative to 11 Labs. Voice Studio is an open-source, fully local alternative

03:09 to 11 Labs for voice cloning, voice design, dubbing, dictation, transcription, and audiobook creation across 646 languages. Everything runs on your machine with no accounts, API keys, or cloud. A model catalog lets you switch between local TTS and speechtoext engines, routing to CPU or GPU. An audiobook editor turns text, EPUB, or PDF into chaptered files with a multivvoice cast. It runs on Mac OS, Windows, and Linux. Download it and

03:39 clone a voice. Project number five, Octtop, self-hosted, multi-user, multi-agent AI assistant. Octtop from Tencent Cloud is an open-source self-hosted AI assistant for multiple users and agents. It runs as a process on your machine serving a web dashboard, a CLI, cron automation, and channels like thaw, ding-talk, discord, and we each user gets a team of agents to switch per task, each with its own workspace. Extend it with an expert

04:09 library, MCP connectors, and ACP for IDE tools with per user isolation and data local. Deploy it and run your assistant. Project number six, Man Flutter. Google’s SDK for cross-platform apps from one codebase. Flutter is Google’s open- source SDK for building apps for mobile, web, and desktop from a single codebase. You write in Dart using a reactive framework with material and Certino widgets, and a portable C++ engine renders every pixel so your UI looks the same across platforms. Hot

04:41 reload shows code changes in a running app and builds compile to native machine code. One codebase ships to iOS, Android, web, Windows, Mac OS, and Linux. Install it and build an app. Project number seven, Staret Pixel Art Space Station, where real agents work. Staret is an open-source local first desktop harness where AI agents do real work. As a pixel art space station, you create agents and lay them out, and the layout is the workflow. Rooms are capability limited teams. Corridors are

05:12 authorization handoffs. And placing an object grants a capability. Multiple agents run each with its own workspace and permissions. Calling models and tools via open router, anthropic, and MCP. It never simulates revenue, work, or spend only real estate downloaded and run your crew. Project number eight, Work Trunk. Get Work Tree CLI for parallel AI agents. Work trunk is an open-source CLI for git workree management built for running AI agents in parallel. Git work trees give each

05:44 agent its own working directory. So many agents claude codecs work at once without stepping on each other’s changes. Three commands make work trees as easy as branches and quality of life features cover hooks on create and merge. LLM generated commit messages and a one command squash rebase merge cleanup workflow. It adds shell integration so commands can change directories, install it and run agents in parallel. Project number nine, dash and GPUi kit. Rust desktop UI framework

06:15 built on GPUi. GPUi kit is an open-source Rust framework for building cross-platform desktop apps with GPUi, the UI toolkit behind the Z editor. It ships 75 plus components and primitives as a styled UI system with shades and ses y style design and lucid icons. One crate reexports GPUi and components so your app lists a single dependency. It adds web assembly, access kit accessibility, UI integration testing, and an optional JavaScript runtime

06:45 across Mac OS, Linux, and Windows. It powers Longbridge Pro. Add it and build a desktop app. Project number 10, Splash, local inference engine for Apple Silicon. Splash is an open-source local inference engine for Apple Silicon built around the model. It runs coding agents and open AAI or anthropic compatible apps on one Mac. Start a server with one command, then point claude code, codeex, open code, or Hermes at it, or use the built-in web UI. It’s accelerated by -2

07:16 speculative decoding, autopairing a matching draft, and serves GGUF and MLX models, including Qentense, and mixture of experts packages. Install it with Homebrew and run a model locally. Project number 11, One Take. Continuous take motion films. No cuts between scenes. One Take is a Clawed agent skill that makes product launch films, teasers, and feature demos as one continuous take. Every beat grows out of the one before with no cuts. Most AI motion videos read like a slideshow of

07:48 separate scenes. One take instead designs the boundary between beats so something survives each transition and becomes the next shot. Continuity is scored by an oracle. It runs in clawed code or codecs free for non-commercial use. Install it and film your launch. Project number 12, MIMEN’s KCC convert comics and manga for e- in readers. KCC or Kindle comic converter is an open-source app that converts comics and manga for e- in readers like Kindle,

08:18 Coobo, and Remarkable. feed it image folders, CBZ/PPBR archives or PDFs and it outputs EPUB panel view moi kub CBZ or PDFs so pages fill the screen. It handles manga right to left reading and splitting plus per device profiles. Despite the name, it suits any e-reader. It runs on Windows, Mac OS, and Linux with guey and clas. Install it and convert your comics. Project number 13, OpenAI MCP

08:48 extensions. Build native feeling chat GPT plugins on top of MCP. OpenAI MCP extensions is an open-source SDK and spec that adds chat GPT specific capabilities on top of the model context protocol so developers can build plugins that feel like native firstclass chat GPT features on standard MCP. It layers custom file viewers for supported file types in chat form elicitation settings app deep links and UI metadata like thumbnails. You add it to your MCP

09:19 server through TypeScript or Python SDKs working across Chat GPT desktop web and mobile. Install it and extend Chat GPT. Project number 14, Whiteboard open-source canvas where humans and agents design software. Whiteboard is an open- source desktop canvas where humans and AI agents architect software together. It plugs into tools you already use like clawed code and codecs. Ask your agent to review your branch against main and it writes the review onto the canvas sequence and er diagrams

09:50 and notes from its trace. Click any visualization or a trace line to jump to the underlying code. So you trace requirements to implementation. It runs on local checkouts. Download it and design together. Project number 15, Disc Tree. Tree Map to find and clear disk space. Disc tree Tree is an open- source tree map app for finding and removing what fills your disc. Built for Omari. Pointed at your home folder, a disk, or any directory, and it draws a tree map where the biggest files and folders are obvious. Delete inapp items go to the

10:22 trash or recycle bin. And when it finishes, it rescans and shows how much space you freed. Built-in Rust with GPUi. It runs on Linux, Mac OS, and Windows. Install it and reclaim space. Project number 16, Mandadoodle. Handdrawn art and animation written as code. Anadoodle is an open-source agent skill and TypeScript engine that draws art and animation in code, illustrations, loops, stickers, web art, and scored films in dozens of styles. It’s deterministic. The same source

10:53 renders identical pixels anywhere with no image model and no downloaded assets. Styles can film themselves being drawn. Scores are composed in code and one source exports to PNG, GIF, MP4, WebM, or an offline HTML file. Install it in cloud code and draw in code. Project number 17. Naive N0.5 NMLE model for coding and R&D. Naive N0.5 NMLE model for coding and R&D. Naive N0.5 NMLE

11:25 model for coding and N billion parameter mixture of experts model with 15.5 billion active parameters built for coding and AI R&D by naive AI. It drops global attention for a mix of sliding window attention and lightweight deepsek sparse attention with grouped query attention using 4KV groups reaching native 1 million token context. It supports FP8 mix precision inference on FP8 capable NVIDIA GPUs. Built on an open MIMO base, its weights ship on hugging face in FP8 and GGUF under the

11:58 MIT license. Download it and run it. Project number 18. Yep. Fast tiny package manager for the npm registry. Yep. Is a fast tiny open-source package manager for the npm registry written in Typescript. It’s a dropin for everyday npm use. It reads the same npmrc, accepts npm’s command names and flags and keeps a fpm lock lock file. It installs registry workspace, tarball, link, and hosted git dependencies,

12:28 ddupes versions, and works fully offline from a local store and cache. It’s tuned for speed with near instant warm installs across large monorrepos and a pluggable store backend. Install it and manage your packages. Project number 19, blueprint before slash after. Explain UX decisions as a blueprint animation. Blueprint before slafter is a claw design skill that explains UX decisions as one continuous step-by-step blueprint animation. It turns the screen into a scan blueprint and runs in numbered

13:00 steps, one Y each. In redesign mode, it shows before and after. The live UI becomes a blueprint. The parts that change rebuild. Then the new design appears. In explain mode, it annotates a single screen with guides, dimensions, and labels. It’s free for non-commercial use. Install it and explain your design. Project number 20, IMAGV open type decision model for images and text. Emagev is an open Jev style type decision model that also takes images. You feed it a photo plus app state and

13:31 type questions and it returns calibrated probabilities on the options you set with an explicit can’t tell running locally. It’s built for business decisions over the photos, records, and text you already have. Three sizes ship as Laura adapters on a Quenbase for latency, balance, or quality, and run via MLX or PyTorch. Install it and decide from images. Thanks for watching. See you in the next update.