Task Manager Says 80%. Is That Weird For You?
Task Manager pops up: CPU 80%. But is that normal for you? PC Workman learns your personal baseline, then only speaks up when live behaviour genuinely breaks your pattern. Same for RAM, GPU, temps.
Episodes - start at #1
Your Monitor Shows Numbers. Mine Builds Memory.
Most monitors are obsessed with "now". Your PC isn't a screenshot - it's a story. A small engine samples every second, then compresses 60 raw snapshots into one meaningful average.
How I Almost Made Myself Invisible to AI
Midnight: two pages live, Schema.org in the head, Search Console verified - I nailed it. 7:42 AM: the validator found four critical errors. Looked great to humans, invisible to every AI crawler.
The Canvas Arc Style Nobody Uses
Needed a circular progress gauge. Every answer online: "pip install a library." I drew mine with pure Canvas, zero dependencies. The trick: style="arc", the one parameter 99% of devs never touch.
The Trick Is Knowing When NOT to Flush RAM
Your RAM hits 75%. Most apps panic. I wait 30 seconds to make sure it isn't a loading screen, then call three Windows APIs and free ~500 MB in under a second. The hard part: when NOT to fire.
Nine Layers Decide - Logic or AI
Most AI assistants send everything to ChatGPT. hck_GPT routes every message through a 9-step decision waterfall before it answers - plus intent-aware temperature: 20 settings, not one.
My AI Forgets You on Purpose
Most assistants are digital hoarders. hck_GPT keeps almost no memory - the chat lives in a dict and dies on close. Sounds like a bug; it's the best decision I made. Every answer rebuilt from live data.
A Help Tour That Punches Holes in the Screen
Most apps have a Help page. Mine dims the whole app, then uses the Windows -transparentcolor trick to cut a glowing hole around the exact widget it's explaining. Pure Tkinter Canvas.
20 Temperatures in One Line
Every tutorial uses one temperature. hck_GPT uses 20 — "What CPU do I have?" runs at 0.35 (surgeon), "How are you?" at 0.80 (friend). The whole decision is one dict lookup.
Six Memory Depths, One Dictionary
The answer depends on how far back the AI looks. One dictionary maps each question to a time window — 5 minutes to 7 days — and two lines turn it into the AI's context.
The Bug That Worked Perfectly for 4 Months
snapshot() read live CPU/RAM/GPU flawlessly. Every module called it independently — 19 times per message, no errors, for 4 months. The fix: a 1-second TTL cache. 19 calls become 1.
Five Words That Teach the App What "Normal" Means
v1.7.8's thermal baseline starts with six lines: gaming, idle, light, medium, heavy. Once it knows the context, it learns what normal means inside it — for your machine, doing what it's doing right now.
Two Invisible Bugs: One Python Hid, One That Accused NVIDIA
A quality audit found a missing Optional import that only crashes on Python 3.9-3.13 (not my 3.14, thanks to PEP 749 lazy annotations), and a mini anti-virus whose first act was flagging NVIDIA's driver and python.exe as malware. Both invisible. Both needed someone to go looking.
My AI Had a Brain - It Just Couldn't Reach It
Welford baselines and voltage SPC were running and tested - the chat just never called them. A bare except: pass swallowed a NameError and killed the whole fallback for four months. Nothing ever complained. I only caught it because some answers felt dumber than they should.
Newer editions premiere on LinkedIn first and land here over time.