The Butler's Name Is N.I.G.E.L.

Desk note: this one is real tape. Every quote below came off a Discord conversation on July 21 and 22, 2026, lightly trimmed for print, nothing invented. He is not big on socials, so there is no handle to follow. His words on that: I'm generally not that active on socials. Besides the odd sarcastic comment and reply.

He answers to Mr Baker. Not a stage name, not a brand. For some reason people always seem to call me Mr Baker, he says, and at some point it stuck. The Cookie Monster logo? That, he insists, is just random.

Behind the name is Lee, a senior IT infrastructure engineer three decades into the industry that has also been his hobby since before he finished school. In his early teens he bought a pile of old IBM x86 workstations from a flea market and built one good machine out of all the parts, so he could install MS-DOS and play the first SimCity. Today the machine on the end of his desk is a 10-inch rack full of Proxmox nodes, and the thing running on it is a household staff of AI butlers with names like N.I.G.E.L. and I.R.I.S.

The origin point is not a startup pitch. It is a cinema seat in 2008.

The tape

Introduce yourself like we just met at a cookout.

Hey, I'm Lee. I'm a senior IT infrastructure engineer.

What problem kept bothering you enough to build this?

No real problem. I've just wanted my own J.A.R.V.I.S. since watching the first Iron Man movie in 2008. I was in the cinema. It was the concept of a virtual assistant that was actually useful and could actually do real things.

What happened in late 2023 or early 2024 that made J.A.R.V.I.S. feel technically possible instead of fictional?

In 2023 ElevenLabs launched. And in 2024 OpenAI released voice mode. They were both so much better and more natural than anything I had seen and used before. By then it became clear that soon we would be using AI to build stuff.

Why couldn't you leave the idea alone?

After 30 years working in IT infrastructure I see the potential for connecting it to literally EVERYTHING.

What is your assistant called, and what can it do?

I have many agent personas. The main two are N.I.G.E.L., Neural Intelligence and Generative Experience Layer, and I.R.I.S., Intelligent Reasoning and Inference System. They are male and female versions of the old English butler style personas, J.A.R.V.I.S. clones. They both share a Honcho workspace so they know all the same facts and memory. I also have a dedicated cybersecurity persona called S.E.N.T.I.N.E.L.: Secure Engine for Networked Threat Identification, Neutralisation, Evaluation, and Learning.

Other random personas I use for testing different agent harnesses and plugins: K.E.V.I.N., K.A.R.E.N., S.T.E.V.E., D.U.M.B.O., and S.T.U.P.I.D., the Synthetic Thinking Unit for Processing, Inference and Decisions. Yes, I like giving my agents silly names.

Why butlers, and why build both instead of picking one?

Old English butler persona because that's what J.A.R.V.I.S. was based on. Both male and female versions for the only reason that matters: because I can. I've tried a few local TTS options but I've not found it good enough yet, and I don't want to spend money on an ElevenLabs subscription at the moment.

You want to connect it to everything. What is connected now, and what would you never give it access to?

I still don't trust modern LLMs enough to let them anywhere near my online shopping or banking. But my main personal assistant agents now have full access to my home router, network switches, wireless access points, and homelab servers running Proxmox. They can run security audits, make firewall changes, make routing and switching changes, create VMs and containers on request, run updates, and any other infrastructure task I give them. I keep meaning to connect them to Home Assistant as well, but never seem to get around to it.

I still don't trust modern LLMs enough to let them anywhere near my online shopping or banking. But they have full access to my home router.

What has an agent failed at badly enough that you had to step in?

Nothing immediately springs to mind, but my agents all have a standing order to stop and notify me if they fail at the same task three times in a row.

What is the gap between N.I.G.E.L. and the movie?

For a truly always-on J.A.R.V.I.S. clone you need it to still be useful offline. Local models actually small enough to run on device still need to improve. And hardware prices need to come down.

The workshop

Walk me through the last time you worked on it.

I'm always tweaking it and trying new skills and plugins. The thing on screen is most often some git repo I have just discovered and thought was worth checking out. I have a cron report that runs at 19:00 every day that lists all the GitHub trending repos that are AI related.

Tell me about one repository you recently discovered.

Recently discovered the Ponytail repo. What caught my attention was the idea of stopping unnecessary bloat in AI generated code that we all know is present but don't want to admit. Still using it in my main Hermes agent system at the moment.

You're evaluating the Pi Agent Harness to maybe replace Hermes as your daily driver. What got your attention?

Hermes has done nothing wrong. I still love it. But I am starting to feel like Hermes, much like OpenClaw, is getting bloated with features I will never use. Pi starts with zero. It has read, write, edit, and bash, nothing else. If you want other tools you have to build them, more specifically get your Pi agent to build them. You end up with an agent harness that is customised 100 percent for you.

Take me into the homelab physically.

It's a DeskPi RackMate T2, a 10-inch wide, 12U high desktop rack on the end of my desk in the home office. A Unifi router and switches, all 10GbE networking. Two Minisforum MS-A1 mini workstations, each with 12-core Ryzen 9 7900 CPUs and 96GB RAM, both with Nvidia RTX 5060 Ti 16GB cards on OCuLink adapters, both running Proxmox. Storage is a Synology DS1821+ NAS with eight 3TB drives and two M.2 SSDs for caching, plus a little Beelink ME mini NAS with six 2TB NVMe SSDs that I'm going to put TrueNAS on, probably running Nextcloud to reduce reliance on public cloud storage.

The Home Assistant integration you keep not getting to: what's stopping the guy who automates everything?

Too many things to try and not enough hours in the day. The to-do list only ever seems to get longer.

You use agents at work daily, on real infrastructure. What do they actually do there?

We are mostly using agents to design and build internal business tools and microservices to make people's jobs easier. I've also started using an agent to help me plan, design, and build out new infrastructure. We are using agents as tools to augment existing staff, not replace them with digital employees. And yes, people at work know I have many agents at home.

The personal projects running right now: name them.

The two main projects are simply to keep learning about AI and agents, and to keep tweaking, changing, and improving how my personal assistants work. The third project I have only recently started thinking about is to create an agent workspace similar to how Space Agent works.

What has it cost you?

The last late night session I had was last week, when I started looking into the Pi Agent Harness and how it could replace Hermes as my daily driver in the future. The only thing it has cost me so far is sleep.

If it never works, was it still worth doing?

It already does work, and it's getting better all the time.

The person

Who do you look up to most?

I don't tend to put people on pedestals, but if I had to pick one person in recent history it's hard to ignore Elon Musk. I may not always agree with him, but it's hard to ignore some of the positive things he has been involved in.

Something people would be surprised to know about you?

Not really. What you see is what you get with me. I have been fortunate enough to work in the industry that interests me the most, and it has been my main hobby since before I finished school. I guess most people wouldn't expect me to be into tabletop roleplay games like Dungeons and Dragons.

D&D: DM or player? Is building agent personas just DMing with extra steps?

Mostly a player. Completely offline. Zero AI involvement. It's for when I want to completely disconnect and detox from IT and AI for a few hours.

The man who gave his network a butler, a security guard, and an agent literally named S.T.U.P.I.D. unwinds by rolling physical dice at a table with no screens on it. His last message of the interview, after two days of answers: Done, I think. Haha.