Searchable transcript of Top Open-Source GitHub Projects : Jellyfin, Harper, OpenCut, Axolotl, Tududi & Amicro #278 — ManuAGI - AutoGPT Tutorials (15:10). 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 self-hosted apps to AI coding agents. 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 cutting-edge 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, open-source AI code review agent for Git diffs. Open Code Review is an open-source command-line code review agent from Alibaba. It reads your Git diff and sends changed files to a configurable LLM through an agent with tools, producing structured comments pinned to exact lines.
01:13 Rather than scanning a diff in isolation, the agent reads full files, searches the codebase, and inspects other changed files for context, with each file reviewed in its own parallel worker. Built-in rules per file type cover null risks, thread safety, SQL injection, and XSS, and you can override them per project. It works with OpenAI and Anthropic APIs.
01:33 Install it and review your next change. Project number two, Ru View, Wi-Fi sensing platform turning radio signals into spatial intelligence. Ru View is an open-source Wi-Fi sensing platform that turns ordinary radio signals into spatial intelligence with no cameras or wearables. Every router already floods a room with radio waves, and people disturb those waves as they move or breathe.
01:57 So, Ruuvi captures channel state information from low-cost ESP32 nodes and reads those disturbances. From that, it detects presence through walls and in darkness, estimates breathing and heart rate, recognizes activity and fingerprints rooms. It runs entirely on edge hardware with no cloud, ships pre-trained models, and connects to Home Assistant, Apple Home, Google Home, and Alexa.
02:19 Flash a node and start sensing. Project number three, Harper, offline privacy-first English grammar checker in Rust. Harper is an open-source English grammar checker written in Rust that runs entirely on your machine. Cloud checkers send everything you write to their servers and add network latency to every revision, while heavier local tools need gigabytes of memory and large data sets.
02:41 Harper lints a document in milliseconds with a small fraction of that footprint, and nothing leaves your computer. It plugs into editors through a language server for VS Code, Neovim, Helix, Emacs, and Zed, ships an Obsidian plugin and a Chrome extension, and compiles to WebAssembly for the browser through Harper.js. Install it and write with privacy.
03:03 Project number four, Jellyfin, free software media system you host yourself. Jellyfin is a free software media system that puts you in control of managing and streaming your own media. It's an alternative to proprietary services like Emby and Plex, serving your library from a dedicated server to end user devices through many apps. This repository holds the back-end server and API, written in C# on .NET for full cross-platform support with FFmpeg handling transcoding and a separate web client for the interface.
03:32 There are no premium licenses, paywalled features, or hidden agendas, and the community translates it into many languages. Set up a server and stream your own library. Project number five, Lingbot Map, feedforward 3D model for streaming scene reconstruction. Lingbot Map is an open-source 3D foundation model that reconstructs scenes from streaming video.
03:51 It's geometric context transformer combines coordinate grounding, dense geometric cues, and long-range drift correction in one streaming framework using anchor context, a pose reference window, and trajectory memory. Because it's feedforward with page KV cache attention, it stays stable across sequences beyond 10,000 frames instead of re-optimizing the whole scene.
04:12 You point it at an image folder or a video and get a point cloud in a browser viewer or render a long flight through offline. Weights and demo data ship on Hugging Face. Feed it a walk-through and watch the scene build. Project number six, paper transformers, CPU-GPU hybrid inference and fine-tuning for MoE models. K Transformers is an open-source framework for running and fine-tuning very large language models through CPU-GPU heterogeneous computing.
04:37 Huge mixture of experts models don't fit in consumer GPU memory, so it splits the work hot experts stay on the GPU while the rest run on CPU kernels tuned for Intel AMX and AVX with INT4 and INT8 quantization and NUMA aware memory. It exposes two capabilities, a KT kernel inference path with a Python API that plugs into SG Lang and fine-tuning through Llama Factory.
05:02 It supports Nvidia, AMD, Intel Arc, and Ascend hardware. Install KT kernel and run a bigger model. Project number seven, OpenCut, free open-source video editor for web and desktop. OpenCut is a free open-source video editor for web, desktop, and mobile. Built as an alternative to CapCut, it exists for three reasons. Your videos stay on your device. Basic editing features stay free rather than sitting behind a paywall.
05:28 And the editor stays simple to use. The web app is built with Next JS, a native desktop app uses GPUI, and a platform agnostic Rust core holds the GPU compositor, effects, masks, and WASM bindings shared across both. You run it locally with Bun and Docker or self-host the whole stack. Clone it and start cutting. Project number eight, hiring agent resume to score pipeline with GitHub signal enrichment.
05:52 Hiring agent is an open-source Python pipeline that turns a resume PDF into an explainable evaluation. It converts the PDF to markdown, then uses ginger templates to ask an LLM for each section separately, assembling basics, work, education, skills, projects, and awards into a structured JSON resume. If the resume lists a GitHub profile, it fetches repositories, classifies projects, and picks the strongest contributions.
06:17 An evaluator then scores open-source personal projects, production work, and technical skills with bonuses, deductions, and written evidence. It runs locally with Ollama or through Gemini. Point it at a resume and read the score. Project number nine, O my Pi, terminal coding agent with the IDE wired in. O my Pi run as OMP is an open-source AI coding agent for the terminal built as a fork of Pi.
06:42 It's focus is the harness around the model. Edits anchor to content hashes, so patches land on the first try or get rejected when the file is stale. Reads return summarized snippets, and search runs in process. Beyond editing, it drives a language server for real renames, attaches a debugger, runs persistent Python and JavaScript cells, browses the web, and fans work out to sub agents.
07:04 It connects to over 40 model providers and installs with one command. Install it and code from your terminal. Project number 10, Axolotl, free and open-source framework for fine-tuning LLMs. Axolotl is a free open-source framework that streamlines post-training and fine-tuning of large language models. It replaces scattered scripts with a single YAML file that carries you through dataset preprocessing, training, evaluation, quantization, and inference.
07:31 It handles full fine-tuning, LoRA, QLoRA, quantization-aware training, preference tuning like DPO and KTO, reinforcement learning, and reward modeling across text and multimodal models from the Hugging Face Hub. Data sets load from local disk, Hugging Face, or cloud storage, and training scales from one GPU to multiple nodes through FSDP, DeepSpeed, and Ray.
07:55 Fetch an example and start training. Project number 11, Penako, shared AI canvas for handwriting and diagrams. Penako is an open-source canvas that lets you think with AI outside a chat box. You write a question, equation, or sketch anywhere on a huge canvas and pause, and it reads your marks along with their spatial relationships, then answers beside them.
08:17 Replies arrive as movable drafts you resize, copy, accept, or discard, so they stay separate from your confirmed ink. The browser sends only the relevant crop and geometry to a local node server, which routes the request to an API, the Codex CLI, or the Claude Code CLI. Install it and draw your next problem. Project number 12, Clay Code terminal coding agent with per request model routing.
08:39 Clay Code is an open-source AI coding agent for the terminal that routes each request to the model tier it actually needs. Its hosted router classifies every message across five tiers, escalating when a task turns out harder than it looked, and tool calls, retries, and failovers don't count against your quota. It indexes your project into a code graph, so the agent queries symbols, callers, and blast radius instead of reading whole files.
09:04 After each edit, it runs your type checker and feeds errors back in the same turn. It supports plan mode, sub-agents, MCP, hooks, and skills. Install it and route your next task. Project number 13, User Tour, open-source platform for in-app user onboarding. User Tour is an open-source user onboarding platform for building in-app product tours, checklists, launchers, and surveys.
09:26 It works with any app that runs in a browser across single-page and multi-page setups, so you don't rebuild flows per framework. You target users with custom attributes and tracked events. Then style each flow with your own text, colors, fonts, and themes. Separate environments like production and staging live in one account. Version tracking shows who changed what and analytics report views, completion rates, and drop-off points.
09:52 Self-host it with Docker, deploy it, and guide your new users. Project number 14, Alexandrie, self-hosted knowledge base with extended markdown editing. Alexandrie is a self-hosted open-source knowledge base built around an extended markdown editor. Note apps tend to be too bare or too heavy, so this one sits in between. One Docker Compose command brings up the full stack and you organize notes into workspaces, categories, and nested documents with tags, bookmarks, and Kanban boards.
10:19 The editor adds colored containers, academic blocks, KaTeX math, footnotes, and live preview while full-text search returns ranked snippets. Per-document permissions run five levels. Sign-in works through OIDC and a PWA keeps notes available offline. Run one command and start writing. Project number 15, Homelab, self-hosted visualizer for your home lab network.
10:43 Homelab is a self-hosted infrastructure visualizer that turns your home lab into an interactive network diagram with live status. Instead of drawing every box by hand, it scans your network with Nmap and fills up pending devices queue you approve onto the canvas. And it can import a Zigbee topology straight from Zigbee to MQTT over MQTT. Each node gets its own health check from ping and TCP to HTTP, Prometheus, or a health endpoint, so online and offline services show at a glance.
11:13 A read-only live view shares the canvas and an optional MCP server lets an AI client read and edit it. Deploy it and map your lab. Project number 16, Frontman, AI coding agent that edits from your browser. Frontman is an open-source AI coding agent that lives in your browser instead of your editor. It hooks into your dev server as middleware, so it sees the live DOM, component tree, computed CSS, routes, and server logs rather than just source files.
11:41 You click an element in your running app, describe the change in plain English, and it edits the real source file with instant hot reload. It integrates with Next.js, Astro, and Vite for React, Vue, and Svelte, and runs only in development, so production builds are unchanged. You bring your own model key, install it, and edit what you see. Project number 17, ComPDF self-hosted private document editing and conversion platform.
12:05 ComPDF self-hosted is an open-source document platform you deploy inside your own infrastructure. Rather than integrating an SDK, you run the whole stack with Docker Compose and get a browser tool center that merges, splits, rotates, watermarks, encrypts, and decrypts PDFs. It converts between PDFs, Word, Excel, Slides, HTML, TXT, CSV, RTF, JSON, and many image formats, including searchable PDF and markdown output.
12:34 A dashboard console covers API keys, usage, license status, operation logs, accounts, and branding with a REST API behind it. Files never leave your servers. Clone it and build your document center. Project number 18, Two Due Duty, self-hosted system for organizing tasks and projects. Two Due Duty is an open-source self-hosted app for organizing life and work through a clear hierarchy.
12:56 Tasks belong to projects. Projects group into areas, and notes and tags attach across both, so everything stays connected instead of scattered. It handles subtasks, due dates, and recurring tasks with flexible patterns, custom intervals, and completion-based repeats. Telegram integration captures tasks on the go and sends daily digests. CalDAV syncs with clients like Apple Reminders and Nextcloud, and OIDC brings single sign-on.
13:22 A REST API with personal keys supports automation. Run one Docker command and get organized. Project number 19, Dash-wise, all-in-one self-hosted dashboard for your home lab. Dash-wise is an open-source self-hosted home lab dashboard that gathers your services into one page. Most dashboards expect you to hand-edit config files, so this one puts links, search engines, and wallpapers in a GUI you configure in settings.
14:00 Links group into folders and can be monitored for downtime. Glanceable sit beside the clock and movable widgets show key info. It also pulls RSS news, receives push notifications by URL, and offers a control plus K spotlight search across links and integrations. Built-in auth supports OIDC, and it connects to Jellyfin, Carrel Keep, dash. and bezel. Deploy it and open your home lab.
14:25 Project number 20, a micro-curated library of React micro interactions and transitions. Amicro is an open-source library of React micro interactions and transition components built on motion. Rather than handwriting animation code for every button and control, you browse a live sandbox of smooth hardware-accelerated transitions and copy the generated React, Tailwind, and motion source straight into your project.
14:49 Previews render in grid, list, or matrix layouts with instant scaling, and a dynamic scene system switches between dark and light so the copied code matches. Adding your own transition takes three files. It's built with TypeScript, Tailwind, Lucide icons, and Vite. Clone it and animate your interface. Thanks for watching. See you in the next update.