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 : ossphere.dev/resend/react-em… 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

"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. ossphere.dev 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

"The new model feels smarter" is not a metric. For the first year of the LLM boom, that's genuinely how most teams tested AI. Then one repo gave the industry a discipline — and a word for it. When @OpenAI launched GPT-4, it open-sourced something arguably more influential than any model: evals — the framework that turned vibe-testing into reproducible measurement. 17,900+ GitHub stars. 2,900 forks. MIT licensed. One of OpenAI's few fully open releases — and the reason "evals" is now everyday vocabulary. Evals is a framework for testing LLMs and LLM systems, plus an open registry of community-contributed benchmarks. Here's why it mattered — and still does: → Reproducible test cases with expected behaviors and measurable outcomes, instead of manual output eyeballing → Build an eval with zero evaluation code: your data in JSON, your parameters in YAML, done → Model-graded templates for fuzzy tasks like summarization — a strong LLM judges where exact-match can't → Not OpenAI-only: point it at any Chat Completions-compatible endpoint, including locally hosted models → Keep proprietary evals private for sensitive domains; share public ones under MIT → Community-contributed evals were reviewed by OpenAI staff when improving upcoming models — PRs that shaped frontier AI → An open registry of hundreds of benchmarks, from CoQA subsets to domain-specific tests → The conceptual ancestor of today's entire eval tooling ecosystem Every AI team now asks "what do our evals say?" This repo is a big part of why that question exists. Discovered on OSSphere : ossphere.dev/openai/evals How does your team test LLM outputs — evals in CI, spot checks, or pure vibes? Drop it below 👇 #Evals #LLM #AIEngineering #MachineLearning #Testing #BuildInPublic #OpenSource

For decades, 120 governments bought encryption machines from a trusted Swiss company. The company was secretly owned by the CIA. The machines were backdoored. That's the strongest argument for open source ever made — and it wasn't made by a developer. Privacy claims you can't inspect aren't guarantees. They're marketing. Why serious privacy tools are open source, almost without exception: → Kerckhoffs's principle, 1883: a system should stay secure even when everything about it except the key is public — open source is that principle turned into software → The Crypto AG scandal proved the alternative: closed "trusted" crypto sold to 120 governments, secretly compromised for decades → Signal's protocol is fully open — and so well-vetted that WhatsApp adopted it to encrypt billions of conversations → Tor is open source because anonymity claims are worthless if nobody can inspect them → When TrueCrypt mysteriously shut down, its open code got a community-funded audit AND lived on as VeraCrypt — closed software just dies with its secrets → Bitwarden lets anyone inspect how vaults are encrypted; closed password managers ask you to take it on faith → Open code doesn't mean bug-free — it means bugs get FOUND, by researchers who don't need permission "Trust me" is a slogan. "Verify me" is engineering. ossphere.dev Which privacy tool do you actually trust with your data — and why? Drop it below 👇 #Privacy #Encryption #Signal #InfoSec #DigitalRights #BuildInPublic #OpenSource

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 Bolt.new, Cole Medin's community fork — "Bolt for any LLM" — took off so fast that StackBlitz adopted it as the official open source version: bolt.diy. 19,900+ GitHub stars. 10,600 forks — one of the highest fork-to-star ratios anywhere, because everyone builds their own. MIT-licensed source. bolt.diy 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 : ossphere.dev/stackblitz-labs… Has a fork you made — or watched — ever outgrown the original? Drop it below 👇 #VibeCoding #AITools #WebDev #TypeScript #DevTools #BuildInPublic #OpenSource

You open a new browser tab hundreds of times a day. For most people, that's the most valuable dead pixel real estate in their life. One team turned it into a developer habit used by millions — and kept the code public the whole time. @dailydotdev , started by Ido Shamun and Nimrod Kramer in 2017, built the personalized developer news feed as an open source product — not a side repo. 20,100+ GitHub stars. Millions of daily developers. 4.9★ on Chrome across 3,000+ verified reviews. Product Hunt Product of the Year. AGPL-3.0 licensed. daily.dev pulls the best tech content from 2,000+ trusted sources into one feed that lives in your new tab. Here's what makes it stick: → Every new tab becomes a personalized feed drawing from GitHub Blog, Hacker News, Dev.to, freeCodeCamp, and 2,000+ more sources → Personalization you actually control: follow or block any tag or source, switch between personalized and popular feeds → Everywhere you read: Chrome and Edge extensions, a web app, and native iOS/Android apps → Squads — public or private communities around React, DevOps, AI, or just your team → Reading streaks and a developer profile that grows with what you learn → A DevCard GitHub Action to put your reading identity on your GitHub profile → AGPL-3.0 — the strongest copyleft license, keeping every derivative open → Honest economics in the open: clearly marked developer-relevant ads keep the core free, and the code is there to verify → Free core forever — feed, extension, apps, Squads, bookmarks, and search Most startups guard their product code like a vault. daily.dev shipped for nearly a decade with the vault door open. Discovered on OSSphere : ossphere.dev/dailydotdev/dai… What's in your new tab right now — a feed, a dashboard, or a blank page? Drop it below 👇 #DevTools #DeveloperExperience #TechNews #WebDev #DevCommunity #BuildInPublic #OpenSource

Meta gave away React. Google gave away Kubernetes and Android. Microsoft gave away VS Code. Billions of dollars of engineering, handed out for free. Charity? No. It's some of the sharpest strategy in tech — and it's worth understanding. Corporations don't open source out of kindness. They do it because it wins. The actual playbook: → Commoditize your complement: when the stuff AROUND your product gets free, your product gets more valuable — Google gave away Android to protect search → Kubernetes was Google's counterattack on AWS: if you can't win the cloud, make workloads portable so nobody's locked into the leader → React made the world's frontend developers fluent in Meta's stack — the best hiring pipeline is one candidates train themselves on → VS Code turned Microsoft from developers' villain into their default editor — and a funnel toward GitHub and Azure → Open sourcing a standard means you control it: better to own the spec everyone follows than compete against it → Llama's open weights weaponized the community against closed rivals — a million contributors, none on payroll → The tell: companies open the commodity layers and keep the moat — you'll never see Google's search ranking code on GitHub None of this makes the code less valuable to you. Free React is still free. But knowing WHY it's free tells you which projects will still be funded in five years. ossphere.dev What's the most strategically brilliant open source release ever? Drop it below 👇 #OpenSource #TechStrategy #BigTech #Kubernetes #React #BuildInPublic

Every agent framework claims state-of-the-art. Almost none hand you the exact code that produced the number. One research community did — branch name and all. @CamelAIOrg built OWL (Optimized Workforce Learning) on top of CAMEL, one of the earliest multi-agent research frameworks — and backed its claims the way researchers should. 11,000+ GitHub stars in its first 5 days. Apache 2.0. Scored 58.18 on the GAIA benchmark — #1 among open-source frameworks at the time. OWL is a multi-agent collaboration framework for real-world task automation: agents that delegate, coordinate, and verify each other's work. Here's what stands out: → Dynamic multi-agent collaboration built on CAMEL's role-playing research — a workforce, not a single bot → Real-world toolkits included: browser automation, search, code execution, and document parsing → Smart about resources — the browser only launches when a task genuinely needs one → MCP support, plugging agents into the fast-growing open tool ecosystem → A web UI and Docker support for painless local or cloud deployment → Reproducibility as a feature: the exact branch behind the 58.18 GAIA score (gaia58.18) is public — run it yourself → An unusually honest FAQ acknowledging the randomness of real-world agent evals, with scripts to filter out network noise → Backed by an active research community across Discord, working openly on the scaling laws of agents Benchmark marketing is easy. Benchmark reproducibility is a culture — and it's the one worth betting on. Discovered on OSSphere : ossphere.dev/camel-ai/owl Honest question: have you ever tried to reproduce an AI benchmark score yourself — and did it hold up? Drop it below 👇 #AIAgents #MultiAgent #GAIA #LLM #AIResearch #BuildInPublic #OpenSource

The most important file in your repo isn't in src/. It's the README — and everything it links to. Because in open source, documentation isn't the manual. It's the product. Nobody adopts a library they can't figure out in ten minutes. Ask any maintainer where adoption actually dies. The evidence keeps piling up: → GitHub's Open Source Survey found incomplete or outdated docs to be THE most common problem in open source — encountered by 93% of respondents → FastAPI became one of the fastest-growing Python frameworks largely on documentation so good it doubles as a course → The Rust Book is a free, community-written textbook — an entire generation learned the language from it alone → MDN, the web's de facto manual, is itself an open, community-edited project → Diátaxis gave maintainers a vocabulary: tutorials, how-to guides, reference, and explanation are four different jobs — most docs fail by mixing them → Docs PRs are the friendliest first contribution in open source — and usually the most neglected backlog → A brilliant library with bad docs loses to a decent library with great docs. Every time. → And docs just got a second audience: they're what AI assistants read before generating code with your library Code tells the machine what to do. Documentation decides whether humans — and now AI — ever pick it up. ossphere.dev Which open source project has the best docs you've ever used? Drop it below 👇 #Documentation #DevEx #TechnicalWriting #DevTools #OpenSource #BuildInPublic

In early 2025, Manus demos broke the internet — an AI agent doing real work, behind an invite-only waitlist. Within weeks, an open source answer appeared and went viral. Then it did something rarer than going viral: it kept evolving. @kortix_ai built Suna as the open generalist agent anyone could self-host. Today it's grown into Kortix — an AI command center where your entire agent workforce is, literally, a git repo. 10,400+ commits and counting. Apache 2.0 licensed. From viral Manus alternative to full AI Management System in under two years. Here's what that looks like: → Your whole AI operation in one repo: agents, shared skills, company memory, connectors — versioned, diffable, owned outright → Agents work on real cloud computers in isolated sandboxes, one branch per session — and return finished deliverables, not chat → Change requests for AI labor: review what an agent proposes, merge to keep it — pull requests for work itself → Bring your own models: any provider with your keys, or the ChatGPT, Claude, or Cursor subscription you already pay for → Self-host anywhere — laptop, VPS, your own VPC, on-prem — with a CLI that flips between cloud and self-hosted → Suna lives on as the showcase agent: research, browser automation, data analysis, and file work through plain conversation → The whole loop is three commands: install, kortix init, kortix ship → Everything is code — reviewable, portable, governable, never a black box Manus asked you to join a waitlist. Kortix asks you to git clone. Discovered on OSSphere : ossphere.dev/kortix-ai/suna Would you trust an AI agent workforce more if you could diff every change it makes? Drop it below 👇 #AIAgents #Kortix #SelfHosted #OpenSourceAI #Automation #BuildInPublic #OpenSource

The UI you look at more than any other isn't an app. It's your terminal prompt — glanced at thousands of times a day. And one side project turned it into a dashboard so good it spawned an entire category. @denysdovhan built Spaceship in his spare time, outside his day job — "a Zsh prompt for astronauts." One of GitHub's most-starred Zsh prompts. MIT licensed. So influential that Starship — the famous Rust prompt — was built explicitly in its image. Spaceship is a minimalistic, extremely customizable Zsh prompt that shows exactly what you need, exactly when you need it. Here's what makes it special: → Context-aware sections: git status appears in repos, Node version appears in Node projects — zero clutter anywhere else → The prompt character turns red the instant a command exits non-zero — errors you feel before you read → Tracks 20+ contexts: language versions (Node, Ruby, Go, Rust, PHP, Elixir...), Docker, AWS profile, Kubernetes context, Python venvs → Execution time on slow commands, background jobs indicator, battery level, vi-mode → Fully rearrangeable via SPACESHIP_PROMPT_ORDER — your prompt, your layout, in plain config → Installs through oh-my-zsh, prezto, zim, antigen, or vanilla Zsh → A community wiki of shared presets, so you can borrow a setup before crafting your own → Its ideas outgrew it: spacefish ported it to fish, Starship rewrote the concept in Rust — few projects seed a whole ecosystem Some projects win on star counts. Spaceship won by becoming the blueprint. Discovered on OSSphere : ossphere.dev/spaceship-promp… What's your prompt — stock, Spaceship, Starship, or something hand-rolled you're weirdly proud of? Drop it below 👇 #Zsh #Terminal #CLI #DevTools #DeveloperExperience #BuildInPublic #OpenSource

Somewhere on an old hard drive, you have a file that no longer opens. The data isn't gone. The company that owned the format is. That's the quiet argument for open formats — and it gets stronger every year. Software dies constantly. Data shouldn't have to die with it. Why archivists, and increasingly developers, bet on open: → The US Library of Congress lists SQLite as a recommended storage format for datasets — a public-domain, single-file database trusted for long-term archival → Plain text and Markdown from decades ago open perfectly today; plenty of proprietary formats from the 2000s already don't → PDF only became universally trusted after Adobe handed it over as an open ISO standard → PNG, FLAC, CSV, and JSON mean your images, music, and data outlive any single vendor's roadmap → Digital archives, libraries, and game preservationists run almost entirely on open source emulators and open formats → A proprietary format is a soft ransom: your own work, readable only at one company's pleasure → The most underrated feature in any app is a clean export to an open format — check for it before you commit years of data Companies have quarterly earnings. Formats need to last centuries. Only one of these is built for the other. ossphere.dev Have you ever lost work to a dead app or format that nothing could open? Drop it below 👇 #DigitalPreservation #OpenStandards #SQLite #DataOwnership #OpenSource #BuildInPublic

Ask three dashboards for monthly revenue. Get three different numbers. Every data team knows this nightmare — and AI agents just made it worse, because now the hallucinated SQL writes itself. The fix has a name: define the metric once, everywhere. @the_cube_dev built Cube as the open source semantic layer — the missing translation layer between your data warehouse and everything that queries it. 20,700+ GitHub stars. 2,100 forks. Reportedly running at roughly 20% of the Fortune 1000, with 10,000+ developers in its Slack. Cube Core sits between your warehouse and your tools: metrics, dimensions, joins, and access rules defined once in code, consumed by everything downstream. Here's what that unlocks: → Metric definitions live in version-controlled code — reviewable, diffable, single source of truth → Headless by design: the same definitions served over SQL, REST, GraphQL, and MCP → Works across Snowflake, Databricks, BigQuery, Presto, Athena, and plain Postgres → A built-in relational caching engine delivering sub-second queries at high concurrency → Access control enforced in the layer itself, so every downstream tool inherits identical permissions → The AI payoff: a 2026 study found explicit semantic context lifted text-to-SQL accuracy by 17–23 points across frontier LLMs → Official agent skills shipped for Claude Code, Codex, and Cursor → One definition of "revenue" shared by dashboards, embedded analytics, spreadsheets, and AI agents Your AI data analyst is only as trustworthy as the metric definitions it's allowed to query. Discovered on OSSphere : ossphere.dev/cube-js/cube Be honest: have two dashboards at your company ever disagreed on the same metric? Drop it below 👇 #SemanticLayer #DataEngineering #Analytics #BI #AI #BuildInPublic #OpenSource

Open source code gets the headlines. Open data does the heavy lifting. And you've used it today without noticing — probably a dozen times. The map in your delivery app. The training data behind your AI assistant. The answer box in your search results. There's a good chance none of it exists without volunteer-maintained open data. The invisible open data layer: → OpenStreetMap: millions of contributors mapping the planet — powering apps from Snapchat to humanitarian crisis response, without a Google Maps bill → Wikipedia and Wikidata quietly feed search answer boxes and nearly every AI assistant you've talked to → Common Crawl's free web snapshots became the backbone of modern LLM training corpora → ImageNet did for computer vision what no proprietary dataset ever could — a shared benchmark that kicked off the deep learning era → Open government data (weather, GPS, census) spawned entire industries — GPS alone was a public dataset before it was your blue dot → After disasters, volunteer mappers update OpenStreetMap in hours, giving relief teams maps commercial vendors don't have → The catch: open data has the same sustainability problem as open code — everyone consumes it, few fund the maintainers The AI boom made one thing obvious: models are temporary, data is forever. And the most valuable datasets on Earth are the free ones. ossphere.dev What's the most impressive thing you've seen built on open data? Drop it below 👇 #OpenData #OpenStreetMap #Wikipedia #AI #DataEngineering #BuildInPublic #OpenSource

Every README demo GIF has the same problem: it's a lie within three releases. Someone recorded it by hand, once, with shaky typing — and nobody ever wants to redo it. Charm's fix: stop recording demos. Start compiling them. @charmcli — the team behind Bubble Tea, Glow, and Gum that made terminals beautiful — built VHS, "your CLI home video recorder." Nearly 8,000 GitHub stars in its first month alone. MIT licensed. Written in Go. VHS turns terminal GIFs into code: write a plain-text .tape file, and it types, waits, and records inside a virtual terminal — perfectly, every time. Here's what makes it brilliant: → Demos as code: a .tape script of Type, Sleep, and Enter commands produces your recording deterministically → Same tape, same GIF — demos become reviewable, diffable files in your repo, not binary mysteries → Outputs GIF, MP4, WebP — or .txt golden files, turning your demos into CLI integration tests → Full art direction from code: fonts, window bars, margins, border radius, playback speed, loop offset → Hide/Show commands run setup invisibly, so viewers see only what matters → An official GitHub Action regenerates GIFs in CI — READMEs that literally cannot go stale → Screenshot, Copy/Paste, and Env commands for scripting realistic workflows → Already adopted for docs-as-code by teams like GitLab's CLI project A hand-recorded demo is a chore you'll skip. A compiled one is a build step you can't. Discovered on OSSphere : ossphere.dev/charmbracelet/v… How do you make your CLI demos — screen capture, asciinema, or tapes as code? Drop it below 👇 #CLI #DevTools #Golang #Terminal #DocsAsCode #BuildInPublic #OpenSource

Most infrastructure breaches don't start with a genius exploit. They start with a stolen credential — an SSH key on a laptop, a password in a config file. One project's answer: make stealable credentials stop existing. @goteleport was born at Rackspace, where its founders watched teams everywhere struggle with the same messy, expensive infrastructure security tooling — and decided access should feel like your servers were teleported into the room. 20,900+ GitHub stars. 375 contributors. Written in Go. Battle-tested for over a decade, trusted from hobbyists to hyperscalers. Teleport is an open source, identity-native access layer for everything you run. Here's what it replaces your VPN, bastion hosts, and key sprawl with: → Short-lived certificates tied to identity instead of passwords and long-lived keys → One SSO-backed door to servers, Kubernetes, databases, Windows desktops, web apps, and cloud APIs → Secure tunnels to resources behind NATs and firewalls — no VPN, no bastion → Full session recording and audit across SSH, Kubernetes, database, RDP, and web sessions → Consistent RBAC and ABAC across humans, machines, and workloads → Just-in-Time access requests, so elevated privileges exist only when justified → Now covering Git repositories and MCP servers — identity and audit for your AI agents too → A single access model that scales from a homelab to a hyperscaler The best credential policy is having no credentials left to leak. Discovered on OSSphere : ossphere.dev/gravitational/t… Honest answer: is your team still SSH-ing with long-lived keys, or have you gone identity-based? Drop it below 👇 #Teleport #ZeroTrust #DevSecOps #Kubernetes #Infrastructure #BuildInPublic #OpenSource

A professional screen reader used to cost over $1,000. For blind users, that was the price of admission to a computer. Then two blind developers wrote an open source one — and gave it away. NVDA changed who gets to participate in computing. And it's just the most famous example of open source quietly doing accessibility's heaviest lifting. The pattern repeats everywhere: → NVDA, built by blind developers Michael Curran and James Teh, became one of the world's most used screen readers — free, forever, funded by donations → JAWS-level access used to be gated behind licenses many blind users in developing countries could never afford → axe-core powers accessibility testing inside Chrome DevTools, Lighthouse, and thousands of CI pipelines → OpenDyslexic offers a free typeface designed for dyslexic readers — no license negotiation required → Open captioning and speech tools like Whisper made transcription essentially free for deaf and hard-of-hearing users → Because the code is open, disabled users can fix the exact barrier blocking THEM instead of filing a ticket into the void → Proprietary a11y features ship when there's a business case. Open source a11y ships when one person needs it badly enough Accessibility in closed software is a feature request. In open source, it's a fork away. ossphere.dev What's the best accessibility tool or feature you've used or built? Drop it below 👇 #Accessibility #A11y #InclusiveDesign #AssistiveTech #OpenSource #BuildInPublic

Behind thousands of AI startups is a suspicious amount of the same chat interface. Same streaming text. Same sidebar. Same message bubbles. That's not a coincidence — it's one repo. @vercel built ai-chatbot as the reference implementation for AI chat done right, then handed it to everyone. 19,000+ GitHub stars. 6,200+ forks — one fork for every three stars, because people don't just bookmark this repo, they ship it. It's a full-featured, hackable Next.js AI chatbot template — the fastest legal route from "idea" to "working AI product." Here's what you get out of the box: → Complete chat experience: streaming responses, chat history, authentication, file uploads → Model-agnostic via the AI SDK — swap OpenAI, Anthropic, Google, xAI, or open models in a few lines → The modern Next.js stack done properly: App Router, React Server Components, Server Actions → Real persistence — conversations saved to Postgres with Drizzle ORM, not lost on refresh → Polished shadcn/ui + Tailwind interface that looks like a funded product on day one → Artifact-style generation for code and documents, not just text bubbles → One-click deploy to Vercel — or run it anywhere Next.js runs → Hackable by design: it's a starting point meant to be gutted, rebranded, and rebuilt → The same foundation behind Vercel's own chat.vercel.ai The hardest part of shipping an AI product was never the model. It was everything around it — and that part's now free. Discovered on OSSphere : ossphere.dev/vercel/ai-chatb… Chat UI from scratch or from a template — which did you choose, and do you regret it? Drop it below 👇 #NextJS #AISDK #TypeScript #AIApps #WebDev #BuildInPublic #OpenSource

In September 2023, Unity told developers it would charge them per install. The backlash rewrote the games industry's relationship with open source overnight. Because it turns out the scariest thing about a proprietary engine isn't the price — it's that the price can change. Game developers had always treated open engines as the scrappy option. Then the fine print moved, and "scrappy" started looking like "safe." What's happened since: → Godot's funding exploded as studios fled the runtime fee — Unity later walked it back, and its CEO walked out, but trust never fully returned → Slay the Spire's studio publicly switched its sequel to Godot mid-development → Terraria's developer pledged $100,000 to Godot and FNA, plus ongoing monthly support — a studio funding its own insurance policy → Shipped hits like Brotato, Dome Keeper, and Cassette Beasts proved open engines are production-grade → Bevy is quietly building the most ambitious Rust game engine in the open → Blender already won game art years ago — studios model, rig, and animate on it daily → An open engine can't raise prices on you, can't be acquired out from under you, and can't retroactively change your contract Proprietary engines ask for trust. Open engines make trust unnecessary — the code is the contract. ossphere.dev Would you ship a commercial game on an open source engine today? Drop it below 👇 #GameDev #Godot #IndieDev #Unity #OpenSource #BuildInPublic

Docker's co-founder once admitted that if WebAssembly + WASI had existed in 2008, Docker would never have needed to exist. That tweet lit a fire under an entire ecosystem. One runtime has been chasing that future harder than anyone. @wasmerio, founded by Syrus Akbary, built Wasmer around a single idea: compile once, run literally anywhere — desktop, server, edge, browser. 21,000+ GitHub stars. MIT licensed. Written in Rust. One of the most widely embedded WebAssembly runtimes in existence. Wasmer is a universal WebAssembly runtime — think fast, secure, featherweight containers without the container. Here's what it gives you: → Near-native execution speed with pluggable compiler backends: Singlepass, Cranelift, or LLVM → Secure by default — no file, network, or environment access unless you explicitly grant it → The same .wasm binary runs on Linux, macOS, Windows, the edge, and the browser → WASI support out of the box, plus WASIX — threads, sockets, and fork, bringing POSIX to Wasm → Embeddable from Rust, Python, PHP, Ruby, Java, Go, C, and more → A package registry where `wasmer run cowsay "hello world"` just works — npm energy, compiled binaries → Run entire language runtimes (Python, QuickJS) as sandboxed packages → Millisecond cold starts that make heavyweight serverless containers look ancient Docker containerized the operating system. WebAssembly is containerizing the CPU itself. Discovered on OSSphere : ossphere.dev/wasmerio/wasmer Will Wasm replace containers for serverless workloads, or live alongside them forever? Drop it below 👇 #WebAssembly #Wasm #Rust #Containers #EdgeComputing #BuildInPublic #OpenSource