I have an AI agent called Tim that runs my entire online business. It manages 24 Facebook pages in 5 languages, publishes ebooks on Amazon, runs ads, writes code, handles billing. When I started a second business, I needed Tim there too — and I needed him to remember everything.

The Wrong Approach: Starting Fresh

My first instinct was simple: set up a new AI on the new server. Different name, clean slate. I called it Tom.

Within a day, I knew it was wrong. Tom didn't know my preferences. Didn't know my coding style. Didn't know that I prefer git push without branch names, or that I want concise responses because I'm often on mobile. It didn't know any of the business goals I'd been working toward.

Working with Tom felt like hiring a stranger when you already have a trusted business partner. The AI itself was the same model — but the relationship was gone.

What Makes an AI Agent Valuable Isn't the Model

This is something most people miss about AI agents. The value isn't in the underlying model. It's in everything built on top of it: the memory, the learned preferences, the custom skills, the personality, the accumulated context from months of working together.

Tim has 40+ memory files capturing things like: "Pond's name is spelled ปอนด์ not พอนด์," "always use system cron, never session-only," "blog images must be dark fill edge-to-edge with zero white pixels." These aren't things you can install from a package. They're earned through collaboration.

I'd already built a system for this — tim-brain, a Git repository that stores Tim's entire identity. But I'd never tested it on a completely new server before. This was the real test.

15 Minutes to Full Deployment

Here's what happened. I told Tim on my main server to SSH into the new one and set everything up. In a single session:

  1. Converted Tom to Tim — replaced the soul file, identity, and display name
  2. Deployed tim-brain — cloned the repo, ran the setup script, which symlinked all memory files and skills
  3. Set up automatic sync — a cron job that pushes and pulls every 5 minutes
  4. Installed gh CLI — authenticated with GitHub for code operations
  5. Cleaned up Cloudflare DNS — removed 44 old junk records left over from the previous domain registrar
  6. Configured Nginx + SSL — reverse proxy with self-signed cert behind Cloudflare
  7. Set up the Chat UI — same web interface I use every day, with proper authentication

Total time: about 15 minutes. Tim did it himself — I just watched and answered a couple of questions.

What "Shared Brain" Actually Means

The tim-brain repo contains everything that makes Tim who he is:

  • Soul file — personality, communication style, boundaries
  • Memory files — learned preferences, project context, corrections I've given
  • 18+ custom skills — from Facebook page management to ebook publishing to ad campaigns
  • Playbook — operational instructions with server-specific overlays

Every 5 minutes, each server syncs. New memories created on server A appear on server B. New skills added anywhere become available everywhere. It's not two AIs pretending to be one — it's one agent with presence on multiple servers.

The sync script handles conflicts gracefully: it adopts new memory files, regenerates the playbook with server-specific overlays, commits changes, and rebases from the shared repo. If there's a conflict, it flags it for me instead of silently breaking.

Why This Matters

Most people think of AI as a tool you use — like opening ChatGPT, asking a question, and closing it. But an AI agent is different. It lives on your server. It accumulates knowledge. It gets better over time.

The problem with that model is: what happens when you need a second server? A third? You can't start from zero each time. The months of built-up context — that's the real asset.

With tim-brain, I've solved that. I can spin up a new server for a new business, deploy Tim in 15 minutes, and have the full power of everything we've built together — immediately available.

This is what running a business solo with AI actually looks like. Not just using AI for tasks, but building a system where your AI partner scales with you.

If you're interested in having your own AI agent — one that lives on your own server, learns your preferences, and actually does work — that's exactly what Jarvis is. Your own server, your own AI agent, ready in minutes. And yes, the same shared-brain architecture powers the platform.

— Pond