You probably have a computer at home that does nothing 23 hours a day. A laptop that's closed. A Mac Mini sitting on a shelf. An old PC under the desk. It draws power, takes up space, and contributes nothing to your life except a Spotify session every now and then.
The base Mac Mini costs around $599. It's silent, draws less power than a lightbulb, and runs whether you're awake or not. That last part is the one that matters. Because the moment you have a machine that's always on, you can hand it the work that's too boring for you to do yourself, and it'll do it forever.
This article shows how to do that with Claude. By the end you'll have three workflows running on your Mac Mini that turn every lecture you watch, every article you save, and every note you ever wrote into a structured second brain that updates itself daily. Total cost: a few dollars a month in API calls. No subscription, no cloud tool, no tab to keep open.
Why your Mac Mini and not the Claude chat?
You can already paste a YouTube link into Claude and ask for a summary. You can already ask it questions about your notes. So why do you need a box on a shelf?
Because every time you do those things in the chat, you are the trigger. You opened the tab. You pasted the link. You typed the question. Close the tab and the work stops. The model is a tool in your hand. The hand has to keep moving or the tool stops.
A Mac Mini doesn't need a hand. You give it a goal once, it runs the cycle on its own, checks its own work, fixes what's weak, repeats until done. Engineers call this a loop, and it's the thing that separates serious AI use from typing into a chat box.
The shape of a loop is simple:
Three parts do the real work, and they're where people get loops wrong. Verify has to be a hard rule, not a vibe, or Claude grades its own homework and writes itself a 9. The loop has to remember what it already tried, or it repeats the same mistake forever. And it has to know when to give up, or it'll run all night and bill you for nothing.
All of that needs somewhere to live. The chat tab is the wrong place. The Mac Mini is the right one.
The stack: Mac Mini, Obsidian, Claude?
Three layers. That's the whole thing.
The box. Mac Mini, always on. Runs cron jobs (launchd on macOS) and a handful of Python scripts. Any cheap mini PC works the same way, the form factor doesn't matter. What matters is the "always on" part.
The store. Obsidian. Free, local, every note is a markdown file in a folder you own. Scripts can read and write to that folder directly. That single fact is what makes everything possible. Your loops drop notes in, your other loops read them back, nothing is locked behind an API.
The brain. Claude API. Sonnet for serious thinking, Haiku for cheap tagging and sanity checks. Pay per use, no subscription. Local Whisper handles transcription so you're not paying per minute of audio.
The glue between them is about 200 lines of Python total. That's the entire setup.
Now the actual loops.
Loop 1: every lecture becomes a note?
You open an Anthropic talk on interpretability. Karpathy posts a 2-hour explainer. Some conference drops a keynote everyone's quoting. You watch one, take no notes, remember almost nothing a week later. The Mac Mini fixes this without you doing anything.
Drop a link, walk away, come back to a structured note that already knows what else in your vault it relates to.
The verify step is the part most people skip, and it's where the whole thing earns its keep. "At least 2 wiki links to existing notes" is a hard rule. Claude can't talk its way past it. Without that rule, the model will happily hand you a vague summary and call it done. With it, the loop is forced to actually connect new ideas to old ones, which is the entire point of having a vault.
Same pattern works on any lecture, any podcast, any conference talk. If it has audio, it becomes a note.
Loop 2: every article you save becomes a note?
You bookmark articles all day. Pocket, Instapaper, Readwise, browser tabs you swear you'll get back to. You never do. The "read later" folder is the place reading goes to die.
The Mac Mini reads them for you.
This loop teaches you the second thing most people get wrong: the stop condition. Without one, the loop will keep retrying forever if Claude struggles with a bad article (paywall, garbage text, scraped wrong), and your token bill will quietly climb past anything reasonable. The rule here is simple. Verify passes, you stop. It doesn't, you skip the article, log why, and move on. The loop never spins on a single bad input.
After a month, your "read later" pile turns into a searchable archive of arguments and quotes, not a graveyard of links you'll never open. Same pattern works on PDFs, newsletters, anything with a text body.
Loop 3: every morning you wake up to a brief?
The Mac Mini is awake before you are. Use it.
This is where the cost question actually matters. Every loop runs on tokens, and tokens are money. The trap is that loops re-read their context on every pass, so a loop that runs ten times doesn't cost ten prompts, it costs ten prompts that each keep getting bigger.
This particular loop dodges that. Haiku is roughly 12x cheaper than Sonnet, the notes are short, the verify is tight, and the whole thing wraps in two or three passes max. Cost per morning: under a cent. Cost per month: less than the coffee you drink while reading the brief.
The principle: use Sonnet only where thinking matters. Tagging, scoring, sanity checks, simple generation, all of that goes to Haiku. The Mac Mini lets you mix models the same way a kitchen lets you mix burners. You don't run the deep fryer to boil an egg.
Try the manual version first?
Before you buy anything or write a single line of Python, you can feel how loops work right now, in the Claude chat, with nothing but a prompt:
Claude drafts, grades itself, finds the weak spot, rewrites, repeats. That's a loop. You built one with a paragraph.
But notice what's missing, because it's the whole point. You're the trigger. Close the tab and it's gone. There's no schedule, no "do this every morning," no waking up to a brief. That gap is exactly what the Mac Mini closes.
The manual version is how you check whether the idea earns its place in your life before you put a machine on it. If a loop is useful enough that you run it three times by hand in the chat, it's worth putting on a schedule. If you don't reach for it the second time, the Mac Mini won't save it.
The order that actually works?
Don't start with the schedule. The people who ship loops that survive in production all build them the same way:
1. Get one manual run reliable, in the Claude chat.
1. Turn that into a Python script on the Mac Mini.
1. Wrap the script in a loop (add the verify gate and the stop condition).
1. Then, and only then, put it on cron.
Skipping ahead, scheduling something you haven't proved by hand, is exactly how loops blow up while you sleep. Prove it once, harden it, then automate it.
What this actually means for you?
The Mac Mini was never the point. The point is having a machine that's always on, so the loops can run without you. The Mac Mini is just the cleanest version of that.
The shift here isn't about a flashier chat app or a new model. It's about who does the work. Claude stops waiting for you to push it through every step and starts running the whole job on its own, on a box you own, writing into files you own, on a schedule you set.
My take: start with the manual loop in this article. Run it in the Claude chat a few times. If you find yourself reaching for it again, plug in the Mac Mini and build the real version. That's the whole stack the serious people use, minus the parts that exist to sell you something.
If you want more breakdowns like this, I post one every couple of days on Telegram and X. Both free.
X — https://x.com/gippp69 Telegram — https://t.me/GipArcAI
