This week I asked my own assistant to tell me everything it could do. For the first time, it gave me the whole answer. Until now it only named about half, so most of what it can do stayed hidden, even from me. That assistant is MakerOS, a private AI assistant for one maker business. It runs on your own computer, under your own Claude subscription, with no third-party server in between: your files and settings stay on your machine. You talk to it in plain sentences. Once you have answered its setup interview, it knows your name, your brand, your voice and your limits, and reads that back every time. A day with it looks something like this. "Give me my morning brief." Your calendar and your inbox, read back with the day's top priority named. "I just finished this piece, give me a caption." One sentence in, and a caption with hook options, a Reel voiceover and a Story prompt comes back. "Help me price this piece." It asks for materials, time and overheads, one at a time, then hands back a breakdown and a suggested price. "Draft my returns policy." A first-draft UK returns policy, built around your own return window, ready for a solicitor to check. It will not send an email, post publicly, spend money on an image, touch your calendar, or delete a file, without asking you first. Day-to-day housekeeping no longer stops to check. Anything that spends or publishes always does. It drafts and it queues. You press the button. This is v1.26.0. It is the release that finally gives you the real answer to what it can do, organised by outcome rather than a partial list. It also checks, on a new machine, for the three things most often missing, and tells you exactly what to fix if one is not there. And a new brand colour palette is now tested against six readability checks before it is used, not just generated and handed over. MakerOS lives in the premium classroom here, as a download. You need the Claude desktop app and your own Claude subscription. Install is one command, then a setup interview: twenty to thirty minutes if you answer everything, nearer ten on the safe defaults. If you want to see what it does with your own brand behind it, that is where to start: https://www.hamedarab.academy/makeros

FreeFlow is my free dictation app for Windows 11, built for jewellers, makers and anyone else on Windows who types all day: emails, listings, quotes, prompts to Claude or ChatGPT. Hold Alt+1 in any app and speak. Release, and the cleaned text lands wherever you were typing, in about two seconds, no separate window, no copy and paste. v2.4.0's one change: the floating gadget is gone by default, replaced by Ghost, a small pill that appears beside your cursor while you dictate and fades once the text is pasted. My own logs made the case for it: 57 days, 2,847 dictations, and the old gadget hidden almost every time I opened it. The gadget is still there, one click away in the tray. Install: 1. Download FreeFlow-Setup.exe from GitHub. 2. Run it. 3. Paste in a free Groq API key when it asks. Windows will flag the installer with a SmartScreen warning because it isn't code-signed yet: that's expected, click More info then Run anyway. The Groq key is free at console.groq.com, no card needed, and covers ordinary daily use on its own. Everything, including screenshots and the source: https://hamedarab.academy/freeflow What do you dictate most, or which app would you most want FreeFlow to behave differently in?

How to change one thing on a design in an AI image tool without it quietly redesigning the whole piece. This came up on the bench again this week, so here is the full method. Keep one reference image as the truth. Pick the single image you are working from and stay on it. Every request you make is a change against that one image, never against a description of it in words and never against a second reference image. The moment a second image enters the conversation, the tool starts blending two aesthetics rather than adjusting one. Name the change as one axis with a number. "Make the panel smaller" is not an instruction, it is an invitation to redesign. The tool has no idea what "smaller" means to you, so it picks. "Panel length down by one third, width unchanged" is an instruction. It names the one thing that moves, the one thing that does not, and by how much. Same piece, one change. Ratios travel as numbers, never as a second picture. If you want a proportion carried across from something else, do not show the tool a second image and hope it understands you mean just the ratio. It will not. It will pull in whatever else is sitting in that second image, usually the ornament and the general look, not only the proportion you wanted. Write the ratio down as a number instead. Generating and editing are two different jobs. If you keep asking for a small change and it keeps coming back as a bigger redesign than you wanted, stop regenerating from the prompt and switch to an edit operation on the actual pixels of the image you already have. An edit modifies what is there. A regeneration makes something new that merely resembles it, and small requests get lost in that gap. Check the thing you did not touch. When you use a reference purely for a proportion, watch that the tool has not quietly dragged the reference's own ornament or detailing back into your piece along with it. Look at the parts you never asked to change, not just the part you did. None of this needs a fancier tool. It is about how you talk to the one you already have.

I removed a screenshot extension from Chrome last week. Not because it was bad, but because I finally read what it could reach: every site I visit, all the time, for a job it only needs to do when I press a button. So I built my own, and it is free for you. https://chromewebstore.google.com/detail/full-page-screenshot/hjkciicndfapkfeaeahnghhkaenfcnam Press the button and it captures the page you are on. If the page is longer than your screen, it scrolls down and photographs the rest, then stitches the whole thing into one image. It also works inside Gmail, where the page itself does not scroll and the message panel does, which is where most screenshot tools give up. What it asks for: the one tab you clicked on, and permission to scroll that tab. Nothing else. No access to other sites, no cookies, no history, and it makes no internet connection at all, so nothing you capture ever leaves your computer. That last part is the reason I am posting it here rather than just using it myself. Checking what a browser extension can reach takes about thirty seconds once you know where to look, and most people never look. Open chrome://extensions, click Details on anything you have installed, and read the Permissions section. If a to-do list wants to read and change your data on all sites, ask why. If you want it to do something it does not do yet, tell me and I will build it.

This week's three. A mirrored twist breaks under the finger. The first CAD came back this week for a commission in progress, and I caught a defect in it. The two halves of a twisted band had been mirrored to each other instead of rotate-copied. A twist is chiral, like a screw thread: mirror it and the direction reverses, so the weave visibly breaks at the join. In flat front and side views it is invisible. You only see it once you rotate the model in 3D. The fix: build one half, then rotate-copy it 180 degrees about the ring's vertical axis. Never mirror. Twisted and braided bands are rotate-copy constructions, full stop. Name the cut or the model decides. Generating a product visual this week, the AI model handed back a ring with a round centre stone. The design, and the caption already written for it, both call for oval. One re-roll with the stone shape named explicitly in the prompt came back correct. Any detail you do not pin by name is a detail the model chooses for you, and stone cut is the first thing it takes liberties with. I stop at two attempts. If the second roll is still wrong, the prompt is wrong, not the model. Sizing is a circumference, not a letter. On the same commission, the ring size was locked this week from a circumference measurement, not from a letter size the client remembered. Letter sizes drift between charts and between countries. Millimetres of circumference do not. The lower shank stays plain by design, which means the ring can be stretched up to roughly two sizes later if it ever needs it. Ornament running all the way round would have removed that option for good. Next week I am back on the Foundations course build. When a CAD comes back from your caster, do you check it yourself in 3D, or approve from the renders?