OSSphere

OSSphere – The fastest way to discover @github OSS, what’s actually worth building on | AI-powered | Discover. Contribute. Dominate.

[OSSphere @ossphere_dev]

It's 2026, and one of the hardest front-end targets is still an email opened in Outlook.

Email HTML never got the modern web. It still runs on nested tables and inline styles, and every client renders it differently. In 2015, a team at Mailjet got tired of fighting it.

They built a markup language that compiles down to all that mess, ran it internally for a year, then open-sourced it in 2016 as @usemjml.

18,300 GitHub stars. 2.1M+ npm downloads a week. MIT licensed.

MJML is a markup language for responsive email. You write clean semantic components, and it compiles them into the table-heavy HTML that email clients actually render.

Here's what it gives you:

→ Components like mj-section, mj-column, mj-text, and mj-button replace hand-built nested tables
→ Output is responsive by default and built for the long tail of email clients, Outlook included
→ A CLI with watch mode, plus a Node.js API for compiling templates in your own pipeline
→ Plugins for VS Code and JetBrains IDEs, plus a free online editor
→ MJML 5 swapped the old minifier and beautifier for htmlnano and cssnano
→ v5 also turned file includes off by default, closing off a whole class of file-access bugs
→ Community ports take it everywhere: MRML in Rust (with a WebAssembly build), Emjay in pure Ruby, and Python bindings
→ Used by Sinch Mailjet, Typeform, Spatie, and GrapesJS
→ Email Love built its Figma-to-email plugin on it
→ v5.4.1 shipped this month

The web moved past table layouts two decades ago. Email never did. MJML just makes sure you never have to write them by hand.

Discovered on OSSphere : https://t.co/NFo0V791XS

What's the worst email client bug you've ever had to work around? Drop it below 👇

#Email #HTML #WebDev #Frontend #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Open source gave us free code. Open data gave that code a world to run on.

The map in your running app, the facts behind a search answer, even the text that trained the first big LLMs. A lot of it comes from datasets volunteers built and gave away.

Here's what the open data commons looks like:

→ 2004: frustrated that the UK's Ordnance Survey didn't freely share its map data, UCL student Steve Coast started OpenStreetMap. It now has 10M+ registered users and ~100K monthly active contributors
→ 2010: no detailed digital map of Port-au-Prince existed. In the month after the Haiti earthquake, about 600 remote volunteers mapped it, and rescue teams relied on it
→ 2012: OSM switched from CC BY-SA to the Open Database License, because licenses written for creative works didn't fit data
→ The entire OSM planet file is free to download, so anyone can self-host a map of Earth
→ 2022: Amazon, Meta, Microsoft, and TomTom launched the Overture Maps Foundation, and its open map data draws heavily on OSM
→ Wikidata holds 120M+ items of structured knowledge, all released under CC0
→ Common Crawl has archived petabytes of the web since 2008. A filtered slice made up 60% of GPT-3's weighted training data
→ Open Food Facts passed 4M products from 150 countries, and 100+ apps have been built on it, including Yuka

Open source asks you to share your code. Open data asks you to share what you know.

https://t.co/3VnJBQ0lfi

Have you ever edited OpenStreetMap, Wikipedia, or Wikidata? What did you add? Drop it below 👇

#OpenData #OpenStreetMap #Wikidata #DataEngineering #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

You'd never ship a backend without tests. Most teams still ship LLM features on vibes.

The prompt looks fine in the playground, then hallucinates for a real user on day two. DeepEval brings the unit-test mindset to that problem.

@confident_ai founders Jeffrey Ip and Kritin Vongthongsri built it as pytest for LLMs, then took the company through Y Combinator's W25 batch.

18,500 GitHub stars. 2,000 forks. ~2.5M PyPI downloads last month. Apache 2.0.

DeepEval is an open-source framework for testing LLM apps, from a single prompt to a multi-step agent, with assertions you can run in CI.

Here's what it gives you:

→ Write LLM test cases like pytest unit tests and run them in your CI pipeline
→ 30+ metrics out of the box: answer relevancy, faithfulness, hallucination, bias, toxicity, JSON correctness, and more
→ G-Eval and DAG metrics for defining your own custom criteria
→ RAG checks like contextual precision and recall, so you test the retriever and not just the final answer
→ Agent metrics for task completion, tool correctness, goal accuracy, and step efficiency
→ Multi-turn metrics for chatbots, like knowledge retention and conversation completeness
→ Synthetic dataset generation for when you don't have labeled data yet
→ Standard benchmarks like MMLU, HumanEval, and GSM8K against any model
→ Works with LangChain, LangGraph, LlamaIndex, CrewAI, Pydantic AI, and the OpenAI Agents SDK, plus local models
→ Now in Python and TypeScript, with voice-agent simulation landing this week

If you wouldn't merge untested code, why merge an untested prompt?

Discovered on OSSphere : https://t.co/jjLj0LN0Qw

How does your team decide an LLM feature is good enough to ship? Drop it below 👇

#LLM #AIEvals #Python #Testing #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

In 2013, a father reverse-engineered his 4-year-old son's glucose monitor so he could see the numbers while the boy was at daycare.

Ten years later, the FDA cleared an insulin-dosing app that grew out of the movement he helped start.

Patients and parents with type 1 diabetes didn't wait for the medical device industry. They wrote the code themselves, published it, and gave the movement a name: #WeAreNotWaiting.

Here's how open source changed diabetes care:

→ Feb 2013: John Costik taps the Dexcom receiver's data and uploads his son's readings to a Google Docs spreadsheet every five minutes
→ That uploader grew into Nightscout, "CGM in the Cloud," released publicly in 2014 so families could watch glucose from their phones
→ Nov 2013: Nightscout co-founder Lane Desborough coins #WeAreNotWaiting at a Stanford diabetes data meetup
→ Dana Lewis's glucose alarms were too quiet to wake her. She and Scott Leibrand built louder ones, then a system that could act on the data
→ Feb 2015: OpenAPS publishes an open-source artificial pancreas that closes the loop between monitor and insulin pump, running on hardware as small as a Raspberry Pi
→ By 2022: 2,700+ known users worldwide and 62 million real-world hours of use
→ Loop, AndroidAPS, and Trio carried the idea to iPhone and Android
→ Jan 23, 2023: the FDA cleared Tidepool Loop, the first open-source automated insulin delivery algorithm to earn US regulatory clearance
→ The American Diabetes Association's 2024 Standards of Care recognized open-source systems as safe and effective with proper education and oversight

Open source usually saves developers time or money. Here, it gave families their nights back.

https://t.co/3VnJBQ0lfi

What's the most meaningful open-source project you know that has nothing to do with shipping software faster? Drop it below 👇

#WeAreNotWaiting #HealthTech #Diabetes #PatientInnovation #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Auth. Database. Background jobs. Email. Deploy. Every full-stack app rebuilds the same plumbing from scratch.

Two brothers decided it should be a config file instead.

Twins Matija and Martin Šošić started @WaspLang in 2020 and took it through Y Combinator's Winter 2021 batch.

18,800 GitHub stars. 1,500 forks. MIT licensed. Its Open SaaS starter has another 16,000 stars on its own.

Wasp is a full-stack framework for React, Node.js, and Prisma. You declare what your app is in one spec file, and Wasp's compiler generates the glue.

Here's what it gives you:

→ Declare pages, routes, auth, operations, and jobs in main.wasp.ts, and Wasp writes the plumbing
→ Auth out of the box, email/password and social login, in a few lines of config
→ Type-safe RPC operations: call server functions from React with end-to-end types and no API layer to hand-write
→ Automatic cache invalidation after actions, so the UI stays in sync
→ Background jobs, cron, and email sending built in
→ One command to deploy to https://t.co/T5w24rCAGm or Railway
→ Open SaaS: a free starter with auth, Stripe, Polar, or Lemon Squeezy payments, an admin dashboard, file uploads, and an Astro docs site
→ The high-level spec doubles as guardrails for AI coding agents, with far less boilerplate to get wrong
→ The compiler under the hood is written in Haskell
→ Still pre-1.0 beta, with commits landing every week

Rails gave Ruby developers convention over configuration. Wasp is betting React developers want the same deal.

Discovered on OSSphere : https://t.co/BHZoH7gogv

What's the piece of full-stack boilerplate you're most tired of writing? Drop it below 👇

#React #NodeJS #FullStack #WebDev #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Claude Code, Gemini CLI, and GitHub Copilot CLI all run on a terminal UI library whose layout engine started as a React Native side project in 2014.

They're built on Ink. Ink does its layout with Yoga.

Yoga began life as css-layout, a way for React Native to use Flexbox outside the browser. @MetaOpenSource relaunched it as Yoga in 2016, and it quietly became the layout engine for things that aren't websites.

18,900 GitHub stars. 1,600 forks. MIT licensed.

Yoga is an embeddable layout engine written in C++. Its only job is working out the size and position of boxes, using web-standard Flexbox rules.

Here's where it shows up:

→ It lays out every screen of every React Native app
→ It powers Litho, Meta's declarative UI framework for Android
→ Ink uses it to bring Flexbox to the terminal, which is how Claude Code, Gemini CLI, Copilot CLI, Wrangler, and Shopify CLI get their layouts
→ Vercel's Satori (14K stars) uses it to turn JSX into social preview images
→ Bindings for C++, Java, JavaScript/WebAssembly, and Swift
→ Its tests are generated from HTML fixtures rendered in Chrome, so Yoga's output is checked against a real browser
→ CSS Grid is on the way: Grid types landed in 2026, with an Expo engineer driving the implementation
→ It now lives in the React Foundation's GitHub org, alongside React and React Native
→ Still active, with fixes landing this month

Most developers have never heard of Yoga. Most developers have used something it laid out today.

Discovered on OSSphere : https://t.co/HgHPzQCp2b

Where's the most surprising place you've found web tech running outside the browser? Drop it below 👇

#ReactNative #Flexbox #CSS #CrossPlatform #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

One Neovim repo has more forks than stars. 46,500 forks, 31,500 stars.

That isn't a glitch. It's the whole point.

Dotfiles, the hidden config files behind your shell, editor, and OS, might be the most personal open source anyone ships. Developers publish them, strangers fork them, and good ideas spread one alias at a time.

Here's how dotfiles became their own open-source culture:

→ Hidden files exist by accident. Rob Pike traced them to an early Unix ls shortcut meant to skip . and .., which ended up hiding every file starting with a dot
→ kickstart.nvim is a single, fully documented file built to be forked and rewritten as your own, which is why forks outnumber stars
→ Oh My Zsh turned shared shell config into a community framework: 190K stars, 2,500+ contributors, 300+ plugins, 140+ themes
→ mathiasbynens/dotfiles (31.5K stars, 8.4K forks) made its .macos script a rite of passage for setting sane defaults on a new Mac
→ GitHub Codespaces can auto-install your dotfiles repo into every new codespace, so your setup follows you to any machine
→ Whole tools like chezmoi (21.8K stars) and GNU Stow exist just to keep dotfiles in sync across machines
→ Reading a stranger's dotfiles teaches tricks no tutorial ever covers
→ The catch: dotfiles can run arbitrary scripts, so read before you install
→ And keep secrets out. GitGuardian counted 29 million new hardcoded secrets on public GitHub in 2025

Your dotfiles are a map of every problem you've solved and every shortcut you refuse to live without. Share the map. Keep the keys.

https://t.co/3VnJBQ0lfi

What's the one line in your dotfiles you'd never give up? Drop it below 👇

#Dotfiles #Neovim #Zsh #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Your time tracker knows every app you open, every site you visit, and every minute you step away. So who owns that data?

Erik Bjäreholt used RescueTime for years before summing up the problem in four words: "My data wasn't mine."

So he and his brother Johan built @ActivityWatchIt, a tracker where the answer is always you.

19,000 GitHub stars. 1,000 forks. MPL-2.0. Built by volunteers.

ActivityWatch is a free, open-source, automatic time tracker. It logs what you do on your computer and keeps every byte on your own machine.

Here's what it gives you:

→ Tracks your active app and window title automatically, with no timers to start
→ AFK detection from keyboard and mouse input, so idle time doesn't count as work
→ Chrome and Firefox extensions log the actual tabs and URLs you spend time on
→ Editor plugins track your coding time the way WakaTime does, without the cloud
→ Rule-based categories turn raw window titles into buckets like Work, Comms, or Distraction
→ A timeline view plus a query language for slicing your own history however you want
→ A full REST API (buckets, events, heartbeats) to build your own watchers and dashboards
→ Runs on Windows, macOS, Linux, and Android
→ All data stays local. No account, no cloud upload, no paywall on your own history
→ Still shipping: parity tests for its Rust server rewrite landed this month

Most trackers rent you a report about your own day. ActivityWatch hands you the raw data and the keys.

Discovered on OSSphere : https://t.co/IT646LzEC8

Would you actually want to see where every hour went last week? Drop it below 👇

#Productivity #Privacy #TimeTracking #SelfHosted #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

SQLite is about 156,000 lines of C. Its tests and test scripts add up to about 92 million.

That's 590 lines of test for every line of product, and it's a big reason a database inside nearly every phone and browser almost never makes the news.

In open source, the code gets the stars. The tests are what let a stranger's pull request get merged on a Tuesday afternoon.

Here's what serious open-source testing looks like:

→ SQLite hits 100% branch coverage and 100% MC/DC, the coverage standard used for flight-critical avionics software
→ The twist: SQLite's code is public domain, but two of its four test harnesses (TH3 and dbsqlfuzz) are proprietary. The tests are the moat
→ dbsqlfuzz alone runs about a billion test mutations every day
→ Rust's Crater compiles and tests every crate on https://t.co/a3o0zQXEL9 against the new compiler before a release. The whole ecosystem is the regression suite
→ Google's OSS-Fuzz fuzzes 1,300+ projects at no cost and has helped fix 13,000+ vulnerabilities and 50,000+ bugs
→ GitHub Actions stays free and unlimited on standard runners for public repos, so CI is one of open source's quiet shared resources
→ A strong suite is what lets a maintainer say yes to a stranger's PR without re-reading every line
→ The best first contribution to many projects isn't a feature. It's a failing test that reproduces a real bug
→ Google calls it the Beyoncé Rule: if you depend on a behavior, put a test on it, or accept that someone will break it

Code tells you what a project does today. Tests are the promise it'll still do it tomorrow.

https://t.co/3VnJBQ0T4Q

What's the most impressive test suite you've seen in an open-source project? Drop it below 👇

#Testing #SoftwareEngineering #SQLite #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

He spent four years trying to make his collaborative app scale on traditional databases. Then he stopped trying and built one that lives inside the browser.

That was 2014. @marknadal set out to build a peer-to-peer database that could survive in any browser tab and still sync correctly after going offline.

19,100 GitHub stars. ~48K weekly npm downloads. Your pick of Zlib, MIT, or Apache 2.0.

GUN is an open-source, decentralized graph database. Every browser, phone, and server becomes a peer, and data syncs between them in real time.

Here's what makes it different:

→ ~9KB gzipped, benchmarked by the project at 20M+ API ops/sec
→ One graph model covers key/value, tables, documents, and even video
→ Realtime peer-to-peer sync across browsers, devices, and relay servers
→ Offline-first: writes keep working without a connection and merge when peers reconnect, using CRDT conflict resolution
→ SEA adds end-to-end encryption and user accounts with no central auth server
→ AXE clusters relays into a DHT-style mesh to cut redundant traffic
→ RAD handles storage with a radix-tree serialization layer
→ The Internet Archive's decentralized https://t.co/kq7st2iHDl runs GUN alongside IPFS and WebTorrent
→ Decentralized takes on Zoom, Reddit, Instagram, Slack, and YouTube have been built on it
→ Still shipping: new SEA and AXE fixes landed this month

Firebase rents you a backend. GUN makes every user part of it.

Discovered on OSSphere : https://t.co/CPfcEz119a

Would you trust your app's data to your users' devices instead of a central server? Drop it below 👇

#Decentralized #LocalFirst #JavaScript #Database #P2P #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

An 8-billion-parameter LLM running at 41 tokens per second inside a browser tab. No server. No API key.

Almost every AI chat app sends your prompt to someone else's GPU. WebLLM asks why the GPU already in your laptop can't do the job.

@tqchenml, the CMU professor behind XGBoost and Apache TVM, announced it in April 2023 with a simple pitch: run the whole model in the browser with WebGPU.

19,200 GitHub stars. 1,400 forks. Apache-2.0. New release shipped this month.

WebLLM is an in-browser LLM inference engine. The model downloads once, caches locally, and runs on your own hardware.

Here's what it gives you:

→ Full OpenAI API compatibility: streaming, JSON mode, seeding, logit-level control. Keep your code, drop the server
→ Llama-3.1-8B at 41.1 tokens/sec on an M3 Max, about 71% of native speed
→ Phi-3.5-mini at 71.1 tokens/sec, close to 80% of native
→ WebGPU kernels compiled through MLC-LLM and Apache TVM, with WebAssembly handling the CPU-side work
→ Runs in a Web Worker or Service Worker, so inference never freezes your UI
→ Grammar-based structured generation for JSON that matches your schema
→ Llama, Phi, Gemma, Mistral, and Qwen out of the box, plus any custom model compiled to MLC format
→ Models cache in the Cache API, IndexedDB, or OPFS, so users download them once
→ One npm package or a CDN import to get started
→ WebGPU is now on by default in Chrome, Edge, and Safari 26, with Firefox rolling out

The cheapest inference server is the one your users already own.

Discovered on OSSphere : https://t.co/zL3fxNZESr

Would you ship a feature that runs the LLM on your users' own GPU? Drop it below 👇

#WebGPU #LLM #LocalAI #JavaScript #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Open source gets the stars. Open standards do the heavy lifting.

In 1969, a UCLA grad student named Steve Crocker titled his networking notes "Request for Comments" so they wouldn't sound too official. Nearly 10,000 RFCs later, that humility still runs the internet.

Your browser, your editor, your containers, and your tracing all work across vendors for one reason: somebody wrote the rules down in public, so competing implementations could agree.

Here's how open standards quietly shaped the tools you use:

→ RFC 1 went out on April 7, 1969. The series is now closing in on 10,000 documents
→ The IETF's creed since David Clark's 1992 talk: "rough consensus and running code." No kings, no voting
→ A spec only becomes a full Internet Standard once at least two independent implementations interoperate. The code has to prove the paper
→ In 2004, Apple, Mozilla, and Opera broke from the W3C to write HTML for how the web actually worked. In 2019, the W3C signed on to a single HTML: WHATWG's Living Standard
→ Microsoft, Red Hat, and Codenvy opened up the Language Server Protocol in June 2016. M editors × N languages became M + N, which is why one language server works in VS Code, Neovim, Zed, and Helix
→ Docker donated its container format and runc to the Open Container Initiative in 2015. That's why Podman, containerd, and Kubernetes all run the same images
→ SmartBear handed the Swagger spec to the Linux Foundation in 2015. On January 1, 2016 it became the OpenAPI Specification, and a whole tooling ecosystem grew on top
→ OpenTracing and OpenCensus stopped competing and merged into OpenTelemetry in 2019. It graduated at the CNCF in May 2026 with the second-highest velocity of 240+ projects, behind only Kubernetes

Code wins arguments for a few years. Specs settle them for decades.

https://t.co/3VnJBQ0lfi

Which open standard do you rely on every day without thinking about it? Drop it below 👇

#OpenStandards #IETF #WebStandards #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

A free screen recorder, written in JavaScript, with Vercel, GitHub, and Google on its user list.

In 2016, the remote design collective @wulkanocom got fed up with screen recorders that were too complex, too expensive, or too limited. So they built one the same way they built websites.

19,400 GitHub stars. 900+ forks. 58 contributors. MIT licensed.

Kap is an open-source macOS screen recorder that lives in your menu bar. Click, drag a region, record, export.

Here's what makes it different:

→ Exports the same recording to GIF, MP4, WebM, or APNG
→ Up to 60 FPS, with quality and size controls before you export
→ Optional audio, click highlighting, and built-in trimming
→ Option-click to pause mid-recording
→ UI built with Electron, React, and Next.js, so any web developer can contribute
→ The capture itself runs on Aperture, Wulkano's own Swift library on top of AVFoundation
→ Plugins are plain npm packages that hook the export step: Giphy, Streamable, S3, Vercel, and more
→ Native builds for Apple Silicon and Intel, one brew install --cask kap away
→ The last stable release, v3.6.0, shipped in October 2022. The repo has been quiet since, but it isn't archived

Most screen recorders want a subscription. Kap wants one click in your menu bar.

Discovered on OSSphere : https://t.co/IAmumXHb72

What do you use to record demos and bug repros? Drop it below 👇

#macOS #ScreenRecording #Electron #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

Your code is open source. There's a good chance the font you write it in is too.

Fonts are software. They get versioned, forked, patched, and licensed, and TrueType files even carry bytecode for a tiny virtual machine that snaps each glyph to the pixel grid.

One of the most-used open-source ecosystems on earth ships as .ttf files, and almost nobody talks about it.

Here's what open-source type looks like in 2026:

→ The SIL Open Font License was drafted in 2005 and finalized as v1.1 in 2007. Its signature rule: bundle, modify, and embed the font anywhere, but never sell it on its own
→ Google Fonts hit 1,942 families in August 2026, every one free for commercial use
→ Noto is short for "no tofu," the blank boxes you see when no font has a glyph. It now covers 162 of the 168 scripts in Unicode 16.0 and roughly 1,000 languages
→ Fira Code (82K GitHub stars) built on Mozilla's Fira Mono and turned -> and <= into single ligatures
→ Microsoft released Cascadia Code under the OFL in 2019 for Windows Terminal. JetBrains followed with JetBrains Mono in 2020
→ Nerd Fonts (64.7K stars) patches 10,000+ icons into 67 coding font families. That git-branch icon in your prompt comes from here
→ GitHub's Monaspace is a 5-family monospace superfamily with "texture healing," where narrow letters lend space to wide ones without breaking the grid
→ Red Hat's Liberation fonts match the exact character widths of Arial, Times New Roman, and Courier New, so a Word doc opened on Linux doesn't reflow

We audit our dependencies line by line. Almost nobody checks the license on the font rendering them.

https://t.co/3VnJBQ0lfi

What font is your editor running right now? Drop it below 👇

#Typography #Fonts #DevTools #DeveloperExperience #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

The day OpenAI launched GPT-4, it also open-sourced the tool it used to find GPT-4's mistakes.

March 14, 2023. Alongside the biggest model launch of the year, @OpenAI shipped Evals and asked developers everywhere to go break its model.

The reward for a high-quality eval that got merged? GPT-4 API access. No cash. The repo still collected 1,600+ issues and pull requests.

19,500 GitHub stars. 3,100 forks. MIT licensed.

Evals is a framework for testing LLMs, and the systems built on top of them, against benchmarks you define yourself.

Here's what it gives you:

→ An open registry of ready-made evals probing different dimensions of model behavior
→ Eval templates: write a new benchmark with JSON data and a YAML config, zero code
→ Model-graded evals, where one model scores another model's answers against your rubric
→ A Completion Functions protocol, so you can evaluate prompt chains and tool-using agents, not just a bare model
→ One CLI command to run any registered eval against any model
→ Optional result logging to Snowflake to track runs over time
→ Community evals ranging from everyday reasoning to IMO problems with exact answers
→ OpenAI's hosted Evals dashboard goes read-only Oct 31 and shuts down Nov 30, 2026. This repo isn't affected.
→ MIT license, so the harness and every benchmark in it are yours to fork

OpenAI used the whole internet to find GPT-4's blind spots. The same harness can find your app's.

Discovered on OSSphere : https://t.co/qxe8xxsJf1

What's the one eval you'd write for the model you use every day? Drop it below 👇

#LLM #AIEvals #OpenAI #MachineLearning #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

@ConduitMail @resend Exactly. Sending it is easy. Making sure it actually looks right when someone opens it is the hard part.

[OSSphere @ossphere_dev]

The web abandoned table-based layouts twenty years ago.
Email never got the memo — it's still 1999 in there: nested tables, inline styles, Outlook voodoo.
One team decided developers shouldn't have to time-travel to send a receipt.

@resend — the email platform started by Zeno Rocha — built react-email so you write emails like it's 2026 and let the library speak 1999 for you.

19,600+ GitHub stars. MIT licensed. Years in and still shipping releases near-weekly.

react-email lets you build and send emails using React and TypeScript — components in, client-proof HTML out.

Here's what it handles for you:

→ Write emails as typed React components; it compiles them to the table-based HTML email clients demand
→ A component library that hides the legacy hacks: Button, Section, Row/Column, Img, Markdown, even CodeBlock
→ Tailwind inside your emails — with configs parsed and styles inlined the way Gmail requires
→ A live preview server with hot reload: develop an email like a component, not a mystery
→ Per-client compatibility warnings, filterable to just the clients you actually target
→ Battle-tested details baked in — like padding preview text so Gmail renders it correctly
→ Provider-agnostic sending: Resend, Nodemailer/SMTP, AWS SES, SendGrid, Postmark, anything
→ An open roadmap heading toward a full open source email editor

Every developer eventually learns that email HTML is where CSS goes to suffer. This repo is the exit.

Discovered on OSSphere : https://t.co/iA3xKmvRYj

What's the worst email client rendering bug you've ever had to ship around? Drop it below 👇

#ReactEmail #React #TypeScript #WebDev #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

@i_mika_el @resend Haha yeah, the fact that something this simple needs a hack says a lot about email

[OSSphere @ossphere_dev]

@i_mika_el @stackblitz Exactly! 😄 That fork-to-star ratio really shows how much people are actually building on top of it, not just starring and moving on.

[OSSphere @ossphere_dev]

A company launches a hit AI product. Someone forks it and adds what users were begging for.
Usually this story ends with lawyers.
This one ended with the company making the fork official.

When @stackblitz open-sourced the core of https://t.co/lusjAQaBgG, Cole Medin's community fork — "Bolt for any LLM" — took off so fast that StackBlitz adopted it as the official open source version: https://t.co/xXEk5gG3t4.

19,900+ GitHub stars. 10,600 forks — one of the highest fork-to-star ratios anywhere, because everyone builds their own. MIT-licensed source.

https://t.co/xXEk5gG3t4 lets you prompt, run, edit, and deploy full-stack web apps — entirely in your browser.

Here's what the community built into it:

→ Pick your LLM per prompt: OpenAI, Anthropic, Gemini, Mistral, Groq, DeepSeek, xAI, OpenRouter — or fully local via Ollama and LM Studio
→ A complete dev environment in a browser tab, powered by WebContainers — terminal, filesystem, live preview
→ Preview errors surface as actionable alerts you hand straight back to the AI to fix
→ Attach images to prompts; import existing projects from Git
→ Download any project as a zip, or push deploys straight to Netlify
→ One Docker command to self-host
→ A public roadmap and community governance through the oTTomator Think Tank
→ Hundreds of community contributors shipping features the original never had

Most companies send cease-and-desists to popular forks. StackBlitz sent an org invite.

Discovered on OSSphere : https://t.co/MxEfvtsvp5

Has a fork you made — or watched — ever outgrown the original? Drop it below 👇

#VibeCoding #AITools #WebDev #TypeScript #DevTools #BuildInPublic #OpenSource

[OSSphere @ossphere_dev]

"Open source AI" might be the most abused phrase in tech right now.
Downloadable weights with a restrictive license get called open source daily.
The old guard of open source looked at this and said: not on our watch.

In 2024, the Open Source Initiative — keepers of the definition since 1998 — shipped the Open Source AI Definition. The industry is still fighting about it.

Here's the actual landscape:

→ OSI's bar: freedom to use, study, modify, and share — plus the code, the weights, AND enough detail about training data to meaningfully rebuild the system
→ By that bar, Llama isn't open source: commercial restrictions above 700M users and a usage policy mean it's "open weights," not open source
→ Weights without training data are the new compiled binary: you can run it, even fine-tune it — but you can't truly study or rebuild it
→ Models like AI2's OLMo show the real thing is possible: weights, training data, code, and logs, all published
→ Apache- and MIT-licensed weights (Mistral, DeepSeek) sit in between — genuinely permissive, still not fully transparent
→ Critics call the loose labeling "openwashing": borrowing 25 years of open source goodwill without the obligations
→ Why it matters: research reproducibility, security auditing, and EU AI Act carve-outs all hinge on what legally counts as "open"

Words that took decades to earn their trust can be spent in a press release.
Open source is a definition, not a vibe.

https://t.co/3VnJBQ0lfi

Where's your line — are open weights "open enough," or does open source AI require the data too? Drop it below 👇

#OpenSourceAI #LLM #Llama #AIPolicy #OpenWeights #BuildInPublic #OpenSource