M Asif Rahman

AR.bd is live. Built with AI, and made for AI to read, too. I’ll admit, having my initials on a two-character .bd domain is part of the appeal. Building the website gave me another opportunity to work closely with AI. I used it throughout the process: design, code, content structure, SEO, testing and deployment. I stayed involved in the decisions. We went through revisions, checked the live results, and simplified things whenever the design became more complicated than it needed to be. The site brings together my companies, writing and the AI projects I’m working on. You can explore the project map, read my notes, or ask my AI about my work. There’s also a layer for AI agents: structured public content and an MCP endpoint for finding projects, searching my writing and reading published notes. That felt right for a personal website at this point in my journey. A place people can explore, with information their AI tools can access as well. Visit: https://ar.bd masifrahman.com and asif.im also take you to the same site. I’d love to hear what catches your interest, especially if you find a project you could use. #BuildInPublic #AI

AIScan has now run more than 3,400 scans. The part I did not expect: most of them never came from the website. They came from the API, from MCP, from the command line. I built one scanner, then kept adding different ways to reach it — because the people who care about this problem don't all work in the same place. Seven ways in today: → Website — paste a URL → CLI — npx aiscan-cli amarip.com, 20 seconds, no install → MCP server — 13 tools, so Claude or Cursor can scan a site, read the failing checks and fix them for you → Telegram bot — send it a domain, get the grade back → REST API — OpenAPI 3.1 → Chrome extension → A Claude Code skill Same engine behind all of them. Why any of this matters: an AI agent reads your website nothing like a human does. It never sees your design, your animation, your hero image. It looks for a handful of small files — and most sites have never shipped a single one of them. That is how a genuinely good website ends up invisible to ChatGPT. The free scan stays free. I only charge for the daily monitoring, because that is the part that costs money to run. Use whichever one you already work in. Links in the first comment.

I was checking my Replit profile for something else and noticed a badge I hadn't seen before. 🏆 𝗟𝗲𝘃𝗲𝗹 𝟱, 𝗠𝗮𝘀𝘁𝗲𝗿 𝗕𝘂𝗶𝗹𝗱𝗲𝗿. 𝗧𝗼𝗽 𝟭% 𝗼𝗳 𝗥𝗲𝗽𝗹𝗶𝘁 𝗔𝗴𝗲𝗻𝘁 𝘂𝘀𝗲𝗿𝘀. I didn't know it existed. Nobody sent me an email. It was just sitting there, and I'd apparently earned it at some point without noticing. So I looked up what it actually measures. From under the badge, word for word: 📌 "Operates deployed applications on Replit that serve live traffic and handle requests from real users." Not prompts sent. Not hours inside the tool. 𝗟𝗶𝘃𝗲 𝗮𝗽𝗽𝘀. 𝗥𝗲𝗮𝗹 𝘂𝘀𝗲𝗿𝘀. That's the whole test. That made sense to me, because here's how those apps got there 👇 ━━━━━━━━━━━━━ 🛠️ I started vibe coding in 2024. Replit was my go-to that first year, and I went all in — probably too far in. Over the past two years, across Replit, Claude Code and Lovable, I've built hundreds of small SaaS tools and apps. 𝗡𝗼𝘁 𝗵𝘂𝗻𝗱𝗿𝗲𝗱𝘀 𝗼𝗳 𝘀𝘁𝗮𝗿𝘁𝘂𝗽𝘀. 𝗛𝘂𝗻𝗱𝗿𝗲𝗱𝘀 𝗼𝗳 𝗽𝗿𝗼𝗯𝗹𝗲𝗺𝘀. Almost every one started the same way: ↳ Something in my own work or life was annoying me ↳ Instead of searching for a tool, I built one ↳ Deployed it, and left it running ✈️ One example: visa4.travel — a simple visa-eligibility checker. I travel to a lot of countries, and I was tired of googling the same question every trip. So I built it. Now it answers that question for other people too. I don't build at that pace anymore. But the things I built are still live, still serving people — and that is what the badge was quietly counting. ━━━━━━━━━━━━━ 💡 Which is the only point I want to make today: 𝗜𝘁 𝗱𝗼𝗲𝘀𝗻'𝘁 𝗺𝗮𝘁𝘁𝗲𝗿 𝗶𝗳 𝘆𝗼𝘂'𝗿𝗲 𝘁𝗲𝗰𝗵𝗻𝗶𝗰𝗮𝗹 𝗼𝗿 𝗻𝗼𝘁. 𝗜𝗳 𝘆𝗼𝘂'𝗿𝗲 𝗻𝗼𝘁 𝘀𝗼𝗹𝘃𝗶𝗻𝗴 𝘆𝗼𝘂𝗿 𝗼𝘄𝗻 𝗽𝗿𝗼𝗯𝗹𝗲𝗺𝘀 𝘄𝗶𝘁𝗵 𝗔𝗜, 𝘆𝗼𝘂'𝗿𝗲 𝗽𝗿𝗼𝗯𝗮𝗯𝗹𝘆 𝗻𝗼𝘁 𝘂𝘀𝗶𝗻𝗴 𝗶𝘁 𝗽𝗿𝗼𝗽𝗲𝗿𝗹𝘆. Don't go to AI asking "what can this do?" Go with one annoying problem of your own. Everything else starts from there. 💬 What's the last problem of your own you solved with AI? Even a tiny one counts. #VibeCoding #AI #BuildInPublic #Replit #ClaudeCode #Lovable #SaaS #Founders

My own website scored 75 out of 100 on my own speed scanner. That's a C. 🔎 I posted the scanner yesterday. 486 websites have been scanned since — 721 scans in under a day. Here's what that pile of real sites actually looks like: 🟥 158 F 🟧 149 D 🟨 93 C 🟩 57 B 🟩 20 A ⭐ 9 A+ 63% score below 70. 47% run with no CDN at all. Of the 175 WordPress sites, 82 have no caching plugin installed. Median time to first byte: 223ms. So no — this isn't a post about how everyone else is slow. We're in the C pile with most of you. ⚡ But the number that actually surprised me is 93. 93 of the sites scanned are Next.js. Another 11 Astro, 9 Nuxt, 7 Lovable, 5 Shopify. The web that showed up to be measured is not the web I expected — and most of those people can't install a WordPress plugin to fix any of it. So we did something different with the report. 🤖 Every scan produces two documents, not one. One for you to read. One for your AI agent to read. ↳ Scan your site ↳ Keep the report URL ↳ Paste that URL into Claude, Cursor, Lovable — whatever you build with The agent fetches it and gets the machine-readable version automatically: every failing check, the evidence behind it, the exact fix, and the official guideline the rule comes from. Not a screenshot of a score. An instruction set it can act on. 🎯 And it knows your stack. If it detects Lovable, it won't tell you to set Cache-Control headers — Lovable rewrites those after your worker returns, so that advice would cost you a build cycle for nothing. It tells you what actually works there instead. Same for Next.js, Astro, Nuxt, Shopify. WordPress too, obviously. ✅ Free. No signup. No login. We built it because I got tired of speed reports that hand you a number and leave you to work out the rest. The number is the least useful part — which is lucky, because ours is a C. 👇 Scan your site, hand the URL to your AI assistant, and tell me what it did with it. #xSpeed #WordPress #Scan #PageSpeed #GTMetrix #

Before I told anyone our new speed scanner existed, I pointed it at our own sites. Best score: 83. Worst: 59. That was a humbling morning. 😅 Here's the part that stung. Every one of those sites sits on good infrastructure. TTFB between 154ms and 208ms. Brotli compression everywhere. Cache hit on every single request. The server was never the problem. 📉 One product site: 96 on desktop, 57 on mobile 📉 Another: 92 on desktop, 58 on mobile 📉 Our docs site: 72 on desktop, 43 on mobile — shipping a 7.8MB page We had been reading the desktop number and feeling fine. Our readers were on the other one. ⸻ 🤔 Why did I build this at all? Honestly, laziness. I was going back and forth between Lighthouse and GTmetrix all day to check our own work. Open a tab. Run it. (Or Via MCP, with api limit) Wait. Read a wall of forty audits. Try to work out which three actually mattered. Then do it again on the next site. I had already solved this shape of problem once — I built AIScan.site to check whether a site is readable by AI agents. One URL in, one score out, one clear next step. So I built the speed version. ⸻ ⚡ xSpeed Scan is live, and free 🌐 Any website. Not just WordPress. Not just mine. 🔓 No signup, no email, no credit card. ⏱️ About 40 seconds. 🔗 Every report gets its own shareable URL, desktop and mobile screenshots, a score card you can download, and the full scan history for that domain. But the part I actually care about is what happens after the score. A list of 20 audits is homework. Nobody does homework. ✅ So every failing check comes back with one thing you can act on immediately — the copy-paste prompt that fixes it. Paste it into ChatGPT or Claude and go. 🤖 And if you would rather not be in the middle at all: every report has a machine-readable twin at the same URL. Connect it and your AI agent reads the report, applies the fixes, and re-runs the scan to verify — without a human relaying numbers between two tools. That last bit is why I built this now and not in 2019. The audience for a performance report is not only a developer anymore. ⸻ 👉 Go scan something you own: https://lnkd.in/g_H9ZTjP Then scan a competitor. I won't tell. 😉 And if you want the other half — whether AI agents can actually read your site — that one is at https://aiscan.site Drop your score in the comments. Mine are up there, including the 59. I'll tell you which single fix moves yours the most.

I needed a tool that didn't exist. So I built it in a day, and gave it away. Here's what happened. I run Umami for analytics across 20+ sites. Self-hosted, privacy-first, open source. On xCloud it's a one-click deploy, which is why I never moved to anything else. Yesterday I wanted to stop opening the dashboard. I just wanted to ask: "how did this site do last week, and where did the traffic come from?" For that, a tool needs an MCP server. That's the piece that lets an AI actually use your software instead of describing it. Umami didn't have one. A few people had built community versions. I tried them. Here's the part that cost me an afternoon: They were written for Umami v2. Umami is on v3 now, and v3 renamed things. ↳ They don't crash ↳ They return a clean, confident response ↳ With data for the wrong thing That's a worse failure than an error. An error you notice. A wrong number you repeat in a meeting. So I built my own. One day. And the first thing it told me wasn't a number. It flagged a broken link on my own site. Something had been following it for weeks. No dashboard would ever have shown me that, because it isn't a metric. It's a detail sitting in a list nobody scrolls. Then I looked at what I'd made and felt slightly stupid. Umami is free. Open source. The entire promise is that you own your data. And I'd just built the missing piece for myself and stopped there. So I released it. ↳ 28 tools, covering every report type in Umami v3 ↳ Read-only by default, write access is opt-in ↳ Runs on your machine, credentials never leave it ↳ MIT licensed, free forever One line: npx -y @asif2bd/umami-mcp Here's what I actually took from this. In the agent era, open source has an advantage nobody has priced in yet. When a closed tool is missing something, you file a feature request and wait two quarters. When an open one is missing something, you build it. Then you hand it back, and it's there for everyone who comes after you. Umami had no MCP server on Monday. It has one now. That's the whole difference. Which tool do you use every day that still has no MCP? Name it below. I'm genuinely curious which gap is worth closing next.

AI has made building software dramatically easier. Hosting it is still stuck in the old world. Most cloud platforms ask you to choose a server, stack, and size before you even decide what you want to run. That never made sense to me. With xCloud v2.8.3, we reversed the flow. Start with the application—WordPress, a Git project, an open-source app, or an AI agent. xCloud then identifies compatible servers and guides you toward the right infrastructure when something different is needed. This release also brings: → 32 new One-Click Apps → A completely redesigned Server Overview → Live CPU, RAM, and disk monitoring → A much simpler way to connect OpenAI Codex and Claude Code → Direct environment management for Hermes Agent → 10+ improvements and reliability fixes The feature list is useful. But the direction matters more. 𝗔𝗽𝗽-𝗳𝗶𝗿𝘀𝘁, 𝗻𝗼𝘁 𝘀𝗲𝗿𝘃𝗲𝗿-𝗳𝗶𝗿𝘀𝘁 1/ Pick the app first — WordPress, a Git project, a One-Click App, an AI agent ↳ The workload defines the infrastructure, not the other way around 2/ See which of your existing servers can actually run it ↳ Compatible servers surface first, with their site count for capacity context 3/ Know what you're missing before you provision ↳ Wrong stack or not enough RAM is a message, not a failed deploy 4/ Create a compatible managed server inside the same flow ↳ No bouncing between dashboards 5/ Compare plans, backups and pricing on one screen ↳ Instead of four Small change on paper. It removed the biggest single category of "why did my deploy fail" tickets we had. xCloud started by making WordPress server management easier. Now we’re extending that same simplicity to apps, automations, and AI agents. Build anything. Host anything. By yourself. What should we make one-click next? #SelfHosting #DevOps #CloudHosting #AIAgents

গত পোস্টে জিজ্ঞেস করেছিলাম — কোন market-এর Business Route নিয়ে আলাদা করে লিখব? সবচেয়ে বেশি ভোট পড়েছে 🇨🇦 𝗖𝗮𝗻𝗮𝗱𝗮-তে। গত সপ্তাহেই Toronto থেকে ফিরলাম। তিন বছরে গিয়েছি কয়েকবার — conference, road trip, অসংখ্য founder-এর সঙ্গে বসা। তাই এটা Google করে লেখা না। Disclaimer: আমি visa consultant বা immigration expert নই। একজন founder, যে অনেক founder-কে কাছ থেকে দেখেছে। Founder-দের জন্য আমি মূলত ৪টা পথ দেখেছি: 𝟭. 𝗙𝗼𝘂𝗻𝗱𝗲𝗿 𝗣𝗥 Profile-এর point দিয়ে সরাসরি PR, তারপর সেখানে গিয়েই company দাঁড় করানো। Business এখানে যাওয়ার কারণ না, যাওয়ার পরের কাজ। 𝟮. 𝗦𝗽𝗼𝘂𝘀𝗲 𝗥𝗼𝘂𝘁𝗲 Spouse যান study visa-তে, আপনি dependent হিসেবে। প্রথম বছর similar field-এ job — business culture ভেতর থেকে বোঝার সেরা উপায়। Work hours পূরণ হলে PR apply; অনেক ক্ষেত্রে spouse-এর আগেই আপনার PR আসে। 𝟯. 𝗦𝘁𝗮𝗿𝘁-𝘂𝗽 𝗩𝗶𝘀𝗮 Designated organization-এর support letter লাগে, idea scalable হতে হয়। সবচেয়ে আকর্ষণীয় শোনায় — কিন্তু ওই support letter-ই আসল খেলা। 𝟰. 𝗜𝗖𝗧 — 𝗜𝗻𝘁𝗿𝗮-𝗖𝗼𝗺𝗽𝗮𝗻𝘆 𝗧𝗿𝗮𝗻𝘀𝗳𝗲𝗿 LMIA নেই, IELTS নেই, education requirement নেই — "significant benefit to Canadian economy" category. Process-এ মাসখানেক, day one থেকে PR benefit, এক বছর পর PR file. শর্ত একটাই: 𝗿𝗲𝗮𝗹, 𝗿𝘂𝗻𝗻𝗶𝗻𝗴 𝗯𝘂𝘀𝗶𝗻𝗲𝘀𝘀. এই route-টা যোগ করেছে আমার ছোট ভাই 𝗥𝗶𝗳𝗮𝘁 𝗔𝗵𝗺𝗲𝗱, যার বাসায় Canada-তে থেকেছি। চারটা route আলাদা। গেট একটাই — running business, real revenue, real team। ━━━━━━━━━━━ এবার ৪টা সৎ কথা, যেগুলো কম বলা হয়: 🥶 𝗧𝗵𝗲 𝗰𝗼𝗹𝗱. শীতে -৪০, -৪৫। বের হতে ৩০ মিনিটের প্রস্তুতি, প্রায় astronaut-এর মতো। আমি নিজে শুধু এই কারণেই Canada-তে settle করব না। 💰 𝗩𝗮𝗻𝗰𝗼𝘂𝘃𝗲𝗿-এ ঠান্ডা কম, কিন্তু খরচ অসম্ভব বেশি। 🕌 সন্তান লালন-পালন। বেশিরভাগ province-এর social environment ও school curriculum আমাদের চেনা পরিবেশ থেকে আলাদা। ছোট বাচ্চাদের নিজেদের মূল্যবোধে বড় করতে অনেক পরিবারকেই বাড়তি সচেতন থাকতে হয়। এটা গুজব না — আমি একাধিক পরিবারকে চিনি যারা শুধু এই কারণেই Canada ছেড়েছেন। 🧾 𝗧𝗮𝘅. Canada-র tax resident হলে worldwide income ও সম্পদ report করতে হয় — বাংলাদেশের property, বিদেশে registered company, বিক্রি থেকে আসা লাভ, সবই। যে founder-এর company একদিন অনেক valuable হতে পারে, তার জন্য এটা ছোট বিষয় না। আগে একজন ভালো cross-border accountant-এর সঙ্গে বসুন। ━━━━━━━━━━━ তবে এটা আমার মত, universal সত্য না। আমার পরিচিত অনেক founder সেখানে দারুণ ভালো আছেন — ঠান্ডা তাদের কাছে কিছুই না। আমার কাছে যেটা deal-breaker, আপনার কাছে হয়তো কিছুই না। সিদ্ধান্ত নিন — কিন্তু চোখ খোলা রেখে। 👉 এই ৪টার মধ্যে কোনটা আপনার profile-এর সাথে মিলছে? 𝟭, 𝟮, 𝟯 নাকি 𝟰 — comment-এ শুধু number-টা লিখুন। প্রতিটা পয়েন্টের অনেক বিস্তারিত version লিখেছি আমার Facebook-এ (Asif2BD)। আর কোন অংশটা নিয়ে আরও detail চান — comment-এ বলুন, আমি comment-এই যোগ করে দেব। Next: 𝗗𝗼𝗼𝗿 𝟮 — 𝗧𝗵𝗲 𝗝𝗼𝗯 𝗥𝗼𝘂𝘁𝗲. বাঙালিকে বিশ্বমুখী জাতি হতে হবে। 🌍 ⚠️ UPDATE: উপরের 𝟯 আর প্রযোজ্য না — Start-up Visa বন্ধ। IRCC ডিসেম্বর ২০২৫-এ নতুন application নেওয়া বন্ধ করেছে; শেষ window ৩০ জুন ২০২৬। ধন্যবাদ Nuruzzaman Milon ও Shahnoor Chowdhury Eshan 🙏