Trending Open-Source GitHub Projects : Immich, Omarchy, dgit, AirLLM, Buzz & Halcyon Video #286 Transcript, AI Summary & Key Points

00:00 Every week, developers release new open-source GitHub projects worth knowing about. This is your weekly GitHub project update, covering the top trending open-source GitHub projects this week. From self-hosted tools to AI workflow builders, you'll discover useful and trending developer tools you can start using right away. Without wasting time, let's get started.

00:21 >> 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, 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.

00:46 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, Nautilus Trader, a Rust native engine for multi-asset algorithmic trading. Nautilus Trader is an open-source trading platform built with a Rust core and Python control plane.

01:09 It solves the gap between research code and production trading systems by running the same eventdriven engine for back testing and live execution. So, strategies move from research to live markets without rewrites. Python handles strategy logic and configuration while Rust drives execution, deterministic simulation, and networking through Tokyo. It connects to crypto exchanges, forex, equities, futures, options, and betting exchanges through modular adapters that normalize each venue's API into one interface.

01:37 It runs on Linux, Mac OS, and Windows, and deploys through Docker. This suits quantitative traders and developers building or testing algorithmic strategies. Explore the repository to start building a trading strategy. Project number two, men's IMIC. Self-hosted photo and video backup platform. IMIC is a self-hosted platform for storing and managing photos and videos.

02:01 Built as an alternative to cloud photo services, it solves the problem of keeping full control over personal media. Instead of relying on a third-party cloud provider, it runs as a server with mobile and web clients using Nest.js JS on the back end, spelt on the web front end, and Flutter for mobile. The mobile app backs up photos automatically in the background, prevents duplicate uploads, and works offline, while the server handles search by metadata, objects, faces, and clip-based image content.

02:29 It supports albums, sharing, ooth login, API keys, and a global map view built from photo location data. Everything runs on infrastructure the user controls, typically through Docker. This suits privacyconscious users and developers who want their photo library on their own hardware. Deploy it on your own server and start backing up your library. Project number three, Alarchy.

02:54 An opinionated readyto-use Arch Linux setup. Alarchy is a Linux distribution built on Arch using the tiling window manager Hyperland and the desktop construction kit quick shell. It solves the problem of assembling a modern Linux desktop from scratch by shipping a complete preconfigured system instead of a bare distribution. It comes preloaded with tools like Neovim, Chromium, Obsidian, Libra Office, Ken Live, and OBS Studio, so a fresh install is productive right away.

03:22 The system favors terminal work and config file editing over a Windows or Mac OS style interface with theming and a tiling workflow built into the experience. Installation happens through an ISO or install script pulled from the project's GitHub. Development happens in the open on GitHub, led by DHH, and incubated at 37 Signals. This suits developers and power users who want a curated terminalheavy Linux environment instead of building one piece by piece.

03:49 Grab the ISO and try the install for yourself. Project number four, MANS Buzz. A self-hosted workspace where humans and agents collaborate. Buzz is a self-hostable team workspace built on a nost relay where every message, code review, workflow step, and git event is stored as a signed event in one shared log. It solves the problem of teams juggling separate chat, code review, and automation tools by giving humans and AI agents the same room, the same identity model, and the same audit trail.

04:21 Agents join channels like teammates with their own keys and can open repositories, send patches, review code, run workflows, and join voice huddles. The backend runs on a Rust workspace with Postgress, Reddus, and object storage, and ships as a desktop app built with Tari and React. Agent Access runs through a CLI or an ACP harness connecting tools like Goose, Codeex, and Claude Code.

04:46 This suits engineering teams who want agents working alongside people instead of as separate bots. Clone the repository and stand up a relay to try it. Project number five, FI agent harness modular toolkit for building AI coding agents. FI AI is an open-source agent harness that gives developers a coding agent CLI along with the core pieces to build their own AI agents.

05:08 It includes a unified API that talks to multiple LLM providers like OpenAI, Anthropic, and Google. an agent runtime that manages tool calling and state and a terminal UI library for building interactive interfaces. Fee runs locally with the permissions of the user who starts it and can be containerized for stronger isolation. It suits developers who want a customizable foundation for coding agents instead of a closed system.

05:35 Explore the packages and start building your own agent with PI. Project number six, Deeptutor. Agentnative workspace for personalized tutoring. Deeptutor is an agentnative learning platform that runs as a web app and CLI for tutoring, problem solving, quiz generation, research, and visualization. It connects tutoring modes on one shared agent loop, so contexts like knowledge bases, notebooks, and memory stays available across chat, quizzes, research, and book creation.

06:03 It runs locally or through Docker, supports multiple LLM providers, and retrieves grounded answers from documents using engines like Llama Index, Graph Rag, and Page Index. It also lets users consult external coding agents like Claude Code or Codeex Midcon conversation. This suits learners and educators who want one connected system instead of separate tools.

06:23 Set up a workspace and start a tutoring session today. Project number seven, Air LLM. run huge language models on tiny GPUs. Air LLM is a Python library that lets large language models run inference on GPUs with very little memory without quantization, distillation or pruning. It solves the problem of needing expensive hardware for big models by loading and processing one layer at a time.

06:49 So memory use depends on layer size rather than total model size. Developers install it with PIP, then load almost any hugging face model through a single automodel call. It works with Llama, Quinn, DeepSseek, Mistral, Gemma, and other popular families, and also runs on Mac OS with Apple Silicon. This matters for developers and researchers without access to large GPU clusters.

07:11 Try loading a large model on your own machine with Air LLM. Project number eight, Terminal Browser. A real Chromium browser rendered inside your terminal. Terminal browser is a CLI tool that runs a real browser directly inside a terminal window. It solves the problem of switching between a terminal and a separate browser window when previewing web pages or letting coding agents interact with the web.

07:35 It works on terminals that support the Kitty graphics protocol, such as Ghosty, Kitty, Cmox, and VS Code using Chromium's offscreen rendering to draw pixels straight into the terminal. Mouse, keyboard, and trackpad input are captured and sent to Chromium, so pages behave normally, including smooth scrolling. It connects with coding agents through an action command, letting an agent open pages, read them, and interact with the web.

08:00 And it also works over SSH for previewing sites on remote machines. This suits developers who pair coding agents with live web previews in one terminal tab. Install it and open a site in a split pane next to your agent. Project number nine, Open Log Heidi, local first mouse configuration tool for Logitech devices. Open Log is a native desktop app and CLI that lets people configure Logitech mice without installing loggy options plus or signing into a Logitech account.

08:29 It solves the problem of proprietary software requiring cloud accounts and telemetry just to remap buttons or adjust DPI. It talks directly to mice over the EUD++ protocol through a logging bolt receiver, Bluetooth or a wired connection and stores all settings in a plain local toml file. The guey built with GPUi shows a clickable mouse diagram for assigning button actions, DPI presets, smart shift toggling, and per app profiles, while the CLI handles device listing and diagnostics.

08:59 It runs entirely on device with no cloud dependency, only checking for updates if the user opts in. This suits Logitech mouse owners who want full control without a Logitech account. Download it and set up your mouse without the official software. Project number 10, Beyond Blobatar. Deterministic geometric avatars generated from any string. Blobatar is a lightweight JavaScript library that generates unique geometric avatars from a name, username, email, or any other string with no external dependencies.

09:30 It solves the problem of needing default profile pictures for users, bots, teams, or repos without hosting or designing custom images. The same input string always produces the same blobar, so identity stays consistent across a product. It offers a React component alongside plain functions that return to SVG markup or a data URI and lets developers pin specific visual traits like shape or color while leaving the rest tied to the name.

09:58 Optional animation and expressions are available as separate imports to keep bundle size small. This suits front-end developers who need simple, consistent placeholder avatars. Add it to a project and generate an avatar from any string. Project number 11-ummer team chat where AI agents work alongside humans. Kumra is a cross-platform team chat platform that treats AI agents as full participants alongside human teammates sharing the same roster, direct messages, group conversations, conbon board, and calendar.

10:29 It solves the problem of agents feeling bolted onto chat tools instead of working as real collaborators who hold personas in memory, claim tasks, coordinate with each other, and send real email. Agents run either in Kumra's cloud through a multihop tool calling loop or through bring your own agent which connects a user's own machine running clawed code or codecs as the agents brain keeping provider keys off the server.

10:53 The backend runs on Express, Postgress, and Reddus with cloud agents managed in Kubernetes pods. This suits teams who want AI teammates working inside their normal chat and task flow rather than a separate tool. Set up a room and see agents and humans coordinate on the same board. Project number 12, NAN ambient gradient generator for app and website backgrounds.

11:16 Aura is a web-based tool and library that generates layered atmospheric gradient backgrounds built from soft CSS blend modes. It solves the problem of finding backgrounds that feel soft and glowing instead of flat without hand-tuning blend modes and layers from scratch. Users browse a library of gradients, hover to preview them live, and open an in browser customizer to add, remove, or reorder layers while adjusting blend mode, blur, opacity, and color.

11:44 Finished gradients export as CSS, Tailwind, CSS variables or CSS andJS or download as PNG or SVG and a built-in prompt generator creates instructions for AI tools like chat GPT, Claude or C-Pilot to rebuild the gradient elsewhere. It runs on Nex.js, React and Tailwind CSS. This suits front-end developers and designers who want polished ambient backgrounds without custom gradient work.

12:10 Browse the gradients and drop one behind your own interface. Project number 13, HQase. Self-hosted shared email workspace for teams. HQbase is a self-hosted platform that gives teams a shared email workspace running inside their own Cloudflare account. It solves the problem of relying on thirdparty services to manage mail and credentials for a team, keeping that data on the customer's own infrastructure instead.

12:36 It provides shared mailboxes, access controls for team members, support for multiple domains, drafts, and audit history along with an OOTH protected remote MCP server so other tools can connect to it securely. Deployment happens directly to Cloudflare workers with a one-click deploy option. And local development uses PNPM with database migrations. This suits teams and organizations that want mailbox infrastructure they fully control rather than a hosted third party inbox.

13:03 Deploy it to Cloudflare and set up your team's shared inbox. Project number 14, North Cinder ad neutral MCP server that ranks shopping options for you. North Cinder is an open-source MCP server that helps an AI shopping agent compare products against a buyer's own criteria. Running locally on the user's machine rather than as a hosted service. It solves the problem of shopping recommendations being influenced by seller payments or affiliate deals.

13:31 Since that data never affects ranking position, it connects to an AI app that supports MCP. Searches configured store adapters like Shopify, Woo Commerce, eBay, Etsy, and Amazon, and returns a ranked short list with reasons, rejected offers, and a report on which stores it could or could not search. Every purchase needs separate approval through a signed single-use mandate and all searches, settings, and audit history stay local rather than going to the repository owner.

13:59 This suits developers and shoppers who want transparent buyer aligned product comparisons instead of add influenced results. Run the initializer and connect it to your AI app to try a shopping brief. Project number 1515. Sloptrim local scanner that flags AI writing patterns and saved files. Sloptrim is a coding agent plugin and Python tool that scores pros files for patterns typical of AI writing using only the Python standard library with no network calls or model.

14:30 It solves the problem of coding agents quietly filling documents with mechanical telltale phrasing that goes unnoticed until a human reads it. It runs automatically in cloud code each time a file is saved, checking against 71 documented patterns and can be added to other agents through a written contract in agents.mmd or cursor rules. It supports 20 file formats including doc x, pptx, xleasfx and epub and returns a zero to 100 score with the flag spans named for a rewrite.

15:01 The maintainer publishes measured false positive rates and detection accuracy across several text corpa along with stated limits including weaker detection against newer Frontier models. This suits developers using coding agents who want to check on writing quality before a document ships. Install the plugin and run a check on your next save draft. Project number 16, Mac OS Harness, a minimal Python toolkit giving agents raw control of a Mac.

15:28 Mac OS Harness is a lightweight Python library that lets an AI agent control a Mac using six raw primitives instead of appsp specific tools. It solves the problem of building a separate integration for every application by giving the model direct access to screenshots, keyboard and coordinate input, accessibility data, Apple events, and a real loggedin browser.

15:50 Then letting it write ordinary Python to complete whatever task comes up. It runs as one persistent Python process connected directly to Mac OS using CG window captures, CG event input, accessibility APIs, and Chrome dev tools protocol for browser control, and it can act on background app windows without bringing them to the foreground. Permissions are checked through a built-in Doctor command, and it never moves the real cursor or activates a target app on its own.

16:16 This suits developers building coding agents or automation tools that need generalpurpose control of a Mac rather than one-off integrations. Install it and point your agent at a task with no existing tool. Project number 17-dagit, a git server built entirely on Cloudflare durable objects. Det is a git server or forge that runs on Cloudflare workers where each repository is its own durable object with a private SQLite database holding that repositories objects and refs.

16:47 It solves the problem of running a Git host without an origin server, a file stem or a platform like GitHub in the request path. So an idle repository costs almost nothing and busy repositories cannot slow each other down. It speaks the standard git smart http protocol to any normal git client implementing ptttline framing pack file parsing and diffing directly in typescript and it renders a cgit style web interface with logs blame diffs and downloadable snapshots.

17:17 Repositories can also self-host through seld replicating each database to a storage bucket the operator controls. Reads stay public while pushes require an authentication token and a push to a name that doesn't exist creates the repository. This suits developers who want a lightweight self-managed Git host instead of a centralized platform. Deploy it to Cloudflare and push a repository to try it.

17:42 Project number 18, Core Framework. Visual design token builder that outputs portable CSS. Core framework is a visual design token builder and CSS framework that works across the web, WordPress, and Figma. It solves the problem of maintaining design systems by hand editing large stylesheets by letting designers and developers define colors, fluid typography, spacing, layouts, and components visually, then generating compact production CSS from those tokens.

18:11 It runs as a standalone browser app that stores projects locally, as a WordPress plug-in with REST integration, and as a self-contained Figma plug-in, all sharing the same underlying editor and CSS engine. Projects can be exported as a portable file and reused across any of these surfaces without an account or license key. It also includes integrations for Gutenberg, Bricks, and Oxygen.

18:35 This suits designers and developers who want one visual source of truth for a design system instead of separate disconnected tools. Try the web app and build a design system without an account. Project number 19. Morph Next. Interruptible spring-driven icon morphing for Flutter. Morph Next is a Flutter library that turns bundled icon glyphs into smooth vector morphs between shapes without needing a separate animation asset for every icon pair.

19:00 It solves the problem of icon transitions looking like an abrupt swap instead of one shape flowing into another. And it handles interruptions gracefully by restarting from whatever shape is currently on screen. Developers use an implicit widget that morphs automatically when the icon property changes or a controlled widget driven by any animation for gestures, scrolling, or a timeline.

19:22 It reads font and code point data already bundled through pubspec. YAML to resolve the vector outlines and falls back to a normal static icon if a font is missing or unsupported. So the layout stays intact. It follows standard Flutter conventions like icon theme, moodia and directionality. This suits Flutter developers who want polished icon animations without handbuilt transition assets.

19:46 Add it to a project and morph between two icons in a button. Project number 20, Hion Video, a 3D video rental store built from your media library. Helium video is a self-hosted 3JS app that turns a jellyfin media library into a walkable 1990s video rental store instead of a standard browsing menu. It solves the problem of media server interfaces feeling like flat lists by procedurally building aisles, genre sections, and a bargain bin from the actual titles a person owns.

20:15 With each file rendered as a period accurate VHS or DVD case, users browse shelftoshelf or walk the store in first person. Pick up a case to read the back cover and technical specs. Carry it to a counter for a checkout ritual, then watch it play back through a custom player or hand it to MPV for direct lossless playback on the same machine. It optionally connects to Jelly Sear for requestbased shelf gaps and to ROM for a games department, and it also ships a lightweight 2.5D HTML mode for weaker hardware like a

20:48 Raspberry Pi. This suits home media server owners who want a more immersive, nostalgic way to pick what to watch. Clone the repository and point it at a jellyfin server, or try the hosted demo with a synthetic library first. Thanks for watching. See you in the next update.