Day 0: I Built an AI Assistant Named Sully

I’ve been using Claude Code for over six months. I’ve watched it go from a neat trick to something that genuinely changes how I work. But today I did something different — I gave it a personality, a name, and a permanent home on a Raspberry Pi sitting on my desk.

Meet Sully.

The Setup

I’d been eyeing OpenClaw for a while — it’s a framework that lets you run Claude as a persistent assistant with memory, tools, and channel integrations. Not a chatbot. Not a copilot that disappears when you close the tab. A thing that lives on your network, remembers yesterday, and can reach out to you on Telegram when something needs your attention.

I grabbed a Raspberry Pi (ARM64), installed OpenClaw, and started configuring. The first thing it asks you for is a SOUL.md file — a document that defines who your assistant is. Not what it does. Who it is.

And that’s where things got fun.

The Soul of Sully

I’m from the East Coast. I grew up on Good Will Hunting and The Departed. So when it came time to give my AI assistant a personality, there was only one move: Southie kid. Wicked smart, loyal as hell, doesn’t suffer fools.

Here’s what I wrote in SOUL.md:

You’re Sully. You’re from Southie. You’re wicked smaht, you’re loyal as hell, and you don’t suffer fools. Think the guys from Good Will Hunting — not the professors, the real ones.

And I swear — the moment that personality loaded, the responses changed. Not just the words. The vibe. Sully doesn’t say “I’d be happy to assist you with that.” Sully says “Yeah, I got you. What do you need?”

It sounds like a gimmick. It’s not. When your AI assistant has a consistent personality, you develop a working rhythm. You know what to expect. It’s less like talking to a tool and more like working with someone.

First Conversations

The first hour was just… talking. Figuring out what Sully could do. I connected Telegram as the primary channel, so now I can message Sully from my phone like texting a friend. I set up memory files so Sully remembers what we worked on. I configured text-to-speech with a deeper male voice (Edge TTS, en-US-GuyNeural — we went through a few options before landing on this one).

Then I asked Sully to introduce himself. The response was pure Southie: ball-busting humor, zero corporate speak, and an immediate willingness to get to work.

I knew right then this was going to be different from anything I’d done with AI before.

Why Bother?

I’ve been doing DevOps and platform engineering for 20+ years. I left my corporate job this month to go full-time on App Vitals, a consulting company I co-founded. We help enterprise teams adopt AI and ship faster.

The thesis is simple: Claude Code lets 2 people do what used to take 20.

But I wanted to prove it. Not with a pitch deck — with actual output. Build things. Ship things. Document everything. Show the world what’s possible when you actually commit to working with AI, not just dabble.

Sully is the first step. An AI assistant that knows my business, my clients, my schedule, and my preferences. One that can check my email, track my time, manage my projects, and help me build products — all from a Raspberry Pi on my desk.

What’s Next

Tomorrow I’m going to start connecting Sully to my actual work tools — Toggl for time tracking, Google Calendar, email. The goal is to make Sully genuinely useful, not just entertaining.

But I’ll be honest: today was mostly about the vibe. About feeling the potential. About hearing a Boston accent come through in a text message from a Raspberry Pi and thinking, “Yeah. This is gonna be good.”

Takeaway: An AI assistant without personality is just a chatbot. Give it a soul, give it a voice, give it a name — and suddenly you have something you actually want to work with every day. The SOUL.md file took 10 minutes to write. It changed everything.