Posted 17-May-2026 in Technology

Introducing Igor

We're in an age of personalized software. I see it, I like it, I vibe it, I got it. I've been watching OpenClaw from afar, while forming some ideas around what I'd want from a personal AI assistant.

Keep in mind, I've never actually used OpenClaw. It seemed a bit risky. It also seemed stupid that people were buying Mac Minis to run something that can easily be run on a Linux server. Having an older Intel NUC laying around, I figured that would be more than sufficient to run an agent of my own design. Worst case, I move it to a VPS instance after the proof of concept phase.

A lot of people see "magic" when an agent accomplishes something. I just see a program doing the job it's been designed to do. Sure, the problem solving aspect is easier to implement and more robust with an LLM, but it's still just a program.

Toss it in your crontab (or use systemd) and you got yourself a program that runs on a schedule. Tell the program what you want it to do each time it runs, and you have an agent.

I'm no stranger to scheduled programs kicking major ass for me. I took it a step further recently by automating site discovery on joshing.you. Sites are discovered, work is done to add them to the site, code is committed to a branch, and a merge request is submitted for my approval.

I didn't want any of the work to show up as my user account on my Forgejo server, so I spun up a new user named Igor. Seemed like a good name, and I originally likened the persona to that of "Eye-gore" from Young Frankenstein. That didn't feel quite right, so the perceived persona became that of Tyler, the Creator's IGOR.

Fast forward a few weeks, and I decided to take the next steps and put together a full-fledged coding agent. All very much a work in progress right now, scheduling is nearly locked in. We're working through the hand off of tickets, and what the agent should do in its free time. A lot of effort went into project hand off, and validation before any work happens.

Here's a rough overview of things (some aspirational):

• systemd + shell scripts, because what more do you need?
• Claude Code with an API key rather than trying to piggyback a Claude Max plan
• Forgejo integration via its API
• 6-hour work day, because I'm not a slave driver
• Easy project onboarding, just assign the user to a repo
• Project validation before any work can be done
• Forgejo Issues to triage work, works both ways
• Forgejo Pull Requests for reviewing work
• Weekly maintenance tasks
• Ton of rules and guardrails. No hit pieces from my agent
• Persistent, version controlled memories

And my favorite part of all. I wanted Igor to have a website, so I snagged a domain. Not just any domain, I grabbed Igor.Bot.

As my buddy Geoff said:

can't believe you got that domain you have the craziest luck with domain names

Not wrong, my friend. Not wrong.

So that's the story thus far. I'm still dialing things in, fixing bugs and such. I'm starting to rethink how my projects function to be able to delegate the work better too. Suspect it will take a few weeks to get to a stable point.

Worth noting, while Igor can work on projects, commit memories to its brain, and work on its own website, it can't work on the underlying system. Not because I don't trust, but because I find it to be the equivalent of bathroom surgery.

With Igor being my personal agent, I'm not sure I'll ever open source the code.

:wq