To be completely honest, I don't like my day job. It’s mostly just stress and waiting for the next corporate reprimand. But I can't leave just yet for a few reasons, so I won't get into that.
The real work starts when I get home. I boot up my laptop, load into Fedora 43 (I absolutely love Red Hat-based Linux), and that’s when the David Blaine street magic begins, haha.
The moment I log in, the system starts to choke. A million Docker containers spin up, RAM evaporates like water in a desert, and the CPU gets dangerously close to a kernel panic featuring a very stressed-out penguin. It’s brutal. I launch Devin Desktop, Firefox, and Hermes.
While my agents are writing code and scavenging the web for data to feed their hungry context windows, I hit make ci.
And then... everything dies.
The entire system freezes. The mouse cursor stops moving entirely, like it just saw a cat, haha. Honestly, I freeze along with it, experiencing my own personal kernel panic. 25 minutes of dead silence and staring at a frozen screen later—the agent finally finishes the task.
Here is the biggest lesson I’ve learned about LLMs during these 25-minute freezes: Cheap models are dumb. Expensive models are also dumb, they just forget the conversation slightly later (I’m exaggerating, but you get the point).
My current workflow? I use cheap models like GLM-5.3 Flash for the grunt work. But when it’s time for architectural review, I bring in Fable 5.1 and crank its... let's just say "reasoning dial" to the absolute maximum. Step by step, slowly but surely, my little worker bee agents are building Plënka.
In the next episode of this project, you’ll see—well, read (or have your AI voice assistant summarize for you)—exactly what is inside my Jenkins pipeline. I’ll break down how this strict methodology affects product quality, whether TDD actually helps AI agents, and most importantly, we'll take a look at what exactly is inside my agent.md file.
Drop your thoughts, suggestions, or constructive criticism below. How are you guys surviving local agent workflows without melting your hardware?
Sergei Hanterov
Posted on
The 25-Minute Mouse Freeze: Building an AI Backend on a Dying Laptop
For further actions, you may consider blocking this person and/or reporting abuse
Top comments (1)
Just curious — has anyone here actually managed to run a full agent swarm locally without docker completely eating your RAM?