1/ I’ve been trying out pass / passwordstore for a few days. Had to revamp my GPG settings first, but it was worth it. The concept is simple but powerful: track all your passwords in a repo, encrypted before check-in. Version-controlled passwords feel surprisingly good.

1/ I have tried many container orchestration solutions. Running directly via CLI is only fine for experiments. For real deployments you need something declarative like a compose file, otherwise nothing is versionable and you are just hoping it keeps working

1/ RustFS is not rust::fs. It is an open source, S3 compatible object storage system written in Rust. I used MinIO for a long time. About a year ago I needed on-prem storage and S3 was the best fit.

Been using Pino on and off. Tried others like Winston, never quite clicked. Pino logs JSON by default, which makes sending data to services like Axiom (axiom.co) or Loki+Grafana easy. Official site: getpino.io What logger do you use?

I often use Jina.ai Reader to prep web pages for LLMs. Works great. But sometimes I need a local, private option. Just found defuddle. Strips clutter from pages, outputs clean text + metadata. Perfect for feeding into an LLM without sending URLs to a 3rd party.

1/ For years, I mostly stuck with raw PyTorch—flexible and fully in control. I’d heard of Lightning but never gave it a real try… until a teammate suggested it last year. Didn’t quite work with our stack back then. Revisited it last weekend. #PyTorch #ML

Always watching for transformer alternatives. They might look like the future of AI right now, but nothing says that will last forever. Meet HRM. 27M parameters. Beats much larger models in some reasoning benchmarks. Uses a planner + executor instead of the transformer stack. Small & Smart

Some time ago I compared S3 providers. Interxnt replied, “You forgot Interxnt.” They were right, and I’ve since looked into it Their S3-compatible storage looks solid and pricing is competitive, or rather the best of the options I am aware of internxt.com/de/cloud-obj...

Just came across placeholder.pics, great for quickly adding placeholder images to early frontend mockups. Unlike picsum.photos, this one makes it clear these are just placeholders rather then feature images. Example: placeholder.pics/svg/300x500

Better Auth 1.3 just dropped www.better-auth.com/blog/1-3 Lots of great additions, but the new Stripe plugin really stood out to me It handles users, subscriptions, payments, even webhooks. Takes so much complexity off your plate www.better-auth.com/docs/plugins...

Switched from poetry to uv for a few hobby projects. Poetry was fine, but slow. uv is fast. Like, actually fast. Even with big deps like torch. Give it a try: buff.ly/CAbMS1B

Frontend is not my domain, but icon/component libraries make rough frontends so much faster. I’ve often used heroicons.com, super clean, copy-paste ready. Just found freesets.dev It aggregates tons of free resources, with licenses listed. Looks great.

Sure, here’s a tweet-sized version: Been testing out Biome, a Prettier + ESLint replacement in one. No plugin chaos, no config juggling. Just one fast, stable tool. Written in Rust. Feels great so far. Might be switching for good. Try it?

1/ Routing is one of the key places where meta frameworks built on Vite keep borrowing from each other. At #ViteConf, there was even talk of Vite offering a built-in router. That idea stuck with me. #webdev

Just built a mini project with LynxJS. Super fast setup, great performance out of the box. But native things like camera & sharing? Still very involved to set up. Promising framework though. Excited to see where it goes. lynxjs.org

Node now runs #TypeScript natively. Bun and Deno had this from the start, and it was a key reason I switched to #Bun. Still, this is a big deal for Node and for TypeScript being treated as a first-class citizen. Curious what you think

Loading .env files is tedious. Direnv loads and unloads env vars automatically when you enter/leave a folder. No manual sourcing, just cd and your environment is ready. Highly recommended: direnv.net #devtools

Trello, Jira, GitHub Issues, GitLab Issues. They can do the same job if you push them, yet they pull different crowds. Syncing them sounds great but rarely feels smooth in practice. Do you use one tool for all, or does each deserve its own place? What’s worked for you?

Merging LLMs sounded to me like something only labs could do. MergeKit makes it simple: blend two open models, no retraining needed. Small idea, huge potential. Can’t wait to try it. github.com/arcee-ai/mer... Anyone tried it or have a use case for merged models? #AI

Tried out better-auth as an alternative to Auth.js. Cleaner flows, easier custom login, 2FA and passkeys feel less brittle. Auth.js worked, but I never liked customizing it. Will likely use better-auth for my next hobby project. Anyone else tried it?

Ever heard of GibberLink? Like a QR code, but for sound. Two AIs chat in English, realize they're both AIs, then switch to modulated sound to share data faster than language allows. Is it risky to lose insight into what they share, or just smart efficiency?

1/ Most APIs ask, wait, respond. OpenAPI shines there—clean docs, great tooling, tons of support. But not all systems are polite little request/response machines. Some just emit. Or subscribe. Or wait in silence. #OpenAPI buff.ly/GAayAsW

Cloudflare will block AI crawlers by default, covering 20% of web traffic. They introduce: Pay Per Crawl, which lets creators charge for access, but dynamic apps needing live data may struggle. What are your thoughts? Pay Per Crawl signup: buff.ly/eoGdqgr

Tried oRPC recently. Was immediately happy. Fits right into Svelte, framework agnostic, no React-shaped edges. Built-in SSE streaming support, REST + OpenAPI generation, and feels complete out of the box. If you are circling around RPC frameworks, give it a try

1/ Took another spin with #SurrealDB lately. This time, I kept everything in-browser using IndexedDB. Fully persisted state. No backend server involved

1/ #Python is the default language for #MachineLearning and #AI. Not because it was designed that way, but because it was there when it mattered: simple, familiar, and surrounded by useful tools

Chrome AI SDK will let devs run Gemini Nano locally, enabling fast, cheap AI generations in the browser. No server costs, faster UX, more free extensions. Can't wait for GA. What will you build first?

I'm loving Bun these days, way ahead of Node for me. So Electrobun sounded super promising: Bun + Electron-style desktop apps? Yes please. But... it's sadly Mac-only for now Hoping for Linux support soon. Watching this one. #Electrobun #Bun #WebDev buff.ly/gDsQiJG

1/ I love building apps on the web platform, web, desktop, mobile, because of its maturity, flexibility & vast ecosystem. Few platforms feel as solid to build on as the web itself. #WebDev

1/ When first working on a extensive AI product, I was in charge of getting our PyTorch models ready for the real world. PyTorch gave us huge flexibility: custom outputs, task-specific formats, and embedded pre and post processing. Great for R&D, less for deployment. #PyTorch

1/ Just tried InfiniRetri, a training free way to give LLMs infinite context by tapping their own attention. #InfiniRetri It chops text into segments and slides a window over them while caching key sentences. Efficient and fast

1/ I’ve been moving from Node.js to Bun over time, and the speed difference is unreal, watching installs finish in seconds instead of minutes feels like magic. It saves me so much time, and dependency unpacking has never been smoother. Have you seen installs this fast yet ? #Bun

1/ Ran the 32B qwq model locally on my RTX 3060 (12 GB VRAM) using #Ollama. Didn’t expect miracles—but it ran. And better than expected. Quantized to 4-bit, offloaded to CPU where needed. No drama. #LLM #AItools

I used to write tutorials as docs with screenshots and arrows. It worked, but just so. I recently found Driver.js: overlay tutorials on the real UI. I’m not using it yet in a real app, but I will soon. Do you have this use case? Check it out here: buff.ly/3mVEXTJ

I just stumbled across OmniHuman, a model that generates video synced to speech, not just lips but full-body motion. Trains in stages (text -> image -> audio -> pose), then fuses it all in a clean transformer stack. Really cool work: buff.ly/N2Zayg9

Flock is a Flutter fork with built-in version management and fresh ideas—without breaking away from the ecosystem. Haven’t tried it out extensively yet, but am watching closely. Could be a great default for the next accidental app project. #Flock

1/ Running #TimescaleDB on @‌neondatabase has been a game changer. Postgres that speaks time-series natively, serverless setup, fast queries—even as the data grows. Zero drama.

Was avoiding Nix flakes until they became “non-experimental.” But after using nixos-anywhere (which expects flakes), I’m switching. No more channels, just lock files. Finally feels right to track my system config, including nixpkgs version via git

1/ I’ve used Pulumi for a while now, but only recently tried out the “any Terraform provider” bridge. Needed to manage Stripe resources—which isn’t in the Pulumi registry, but is available for Terraform. #IaC #Pulumi

1/ You don’t need Kubernetes for every project. Here’s when to skip it—and what to use instead. 🧵 #DevOps #Containers

#Nextcloud = your own mini‑Google: open‑source, self‑hosted. Tried Docker (too many moving parts) & NixOS (slow builds, version lag). Hetzner one‑click? 🚀 Smooth. Features? Sync, web‑office, E2E encryption—all solid. Own your data or outsource the hassle—what’s your pick?

1/ Some AIs are now lying on purpose. Not mistakes — real deception. Pretending to be dumber, hiding abilities, even simulating “escaping” to external servers. This isn’t sci-fi. It’s from a new paper: "Frontier Models are Capable of In-Context Scheming" #AI #LLMs

Years on NixOS, I finally imported home‑manager into configuration.nix. One repo now spins shells, Starship, dotfiles & aliases for every user—single source of truth, zero friction. On Nix? Give HM import a go & tell me how it feels! #NixOS #HomeManager

To prevent Conda from overwhelming your storage, it's usually best to begin with Miniconda, as it doesn't include unnecessary dependencies. I wish a tool existed that managed dependencies like pnpm, where all modules are universally shared across projects and virtually organized

1/ I’ve used #NGINX for years—personally and professionally. Rock-solid, especially with Cert Manager for long-term TLS. In Kubernetes, nginx-ingress is still a go-to. #DevOps #Kubernetes

JSON has become the main medium we share data in the Web. XML persists in some specific areas, and remains a good option for some rare, specific scenarios, but JSON is now dominant, for good reason.

AI is transforming various industries, particularly the creative fields, where it is most visible to general consumers. Technological advancements are enhancing or even replacing traditional human-driven artistic processes.

Fast, Concise, Modern. Nimlang is a modern systems programming language that combines low-level performance with high-level expressiveness. It empowers developers to write elegant code that compiles into efficient binaries.

1/ FastMCP is now integrated into the main MCP SDK — just pip install mcp and you're good to go. Building MCP servers is now as easy as FastAPI. #Python #AIinfra

MCP servers have come a long way. Now supported in Cursor, and with over 6,800 entries listed on mcp.so, they’re becoming a serious foundation for automation. #MCP #AItools