For the past year, I've been building an AI agent that lives on my server 24/7. It writes code, manages my Facebook pages, publishes ebooks, runs ads, handles billing — basically runs my entire online business. Every time I posted about it, the same question kept coming up: "Can I have this?"
The Question That Wouldn't Stop
It started as comments on my posts. Then DMs. Then people I know in person asking me the same thing: "How do I get what you have?"
And I'd try to explain — it's not a product, it's my personal setup. It's Claude Code running on a VPS with 18 custom skills, memory files, a soul file, business goals. It's Tim, my AI agent, deeply wired into everything I do.
But that answer didn't satisfy anyone. Because they didn't want to know how it worked — they wanted one of their own.
The Idea: What If I Could Package This?
I spent a few days thinking about it seriously. What would it take to give someone else the same experience I have?
Not a chatbot. Not another ChatGPT wrapper that can write poems and summarize articles. A real AI agent with real server access — one that can install software, write and deploy code, browse the internet, manage files, and actually do work. I wrote a full breakdown of why AI agents work 24/7 while ChatGPT chatbots can't if you want to understand the fundamental difference.
The answer was: give them their own server. Not shared hosting. Not a container in a cluster. A dedicated VPS with their own AI agent installed, configured, and ready to go. Just like mine.
That's Jarvis.
What Every Customer Gets
When someone signs up for Jarvis, here's what happens — automatically, in minutes:
- A dedicated VPS is provisioned from a snapshot I've prepared (Singapore, USA, or EU)
- Claude Code is installed and configured as their AI agent
- A chat interface is deployed so they can talk to their agent from any browser
- A file browser is set up for visual server management
- A custom subdomain is created (e.g.,
jarvis-you.incomeinclick.in.th) - Welcome email with login credentials is sent automatically
No SSH required. No technical setup. They open a browser, start chatting with their AI agent, and it has full access to their server. Want to build a custom tool? Tell the agent. Want to set up a content scheduler? Tell the agent. Want to automate your billing? You get the idea.
Building the Control Plane
The product itself needed infrastructure. I built a complete control plane — the system that manages everything behind the scenes:
- Stripe integration — checkout, webhooks, subscription management, 6 pricing tiers
- Hetzner API — auto-provision and delete VPS servers
- Cloudflare DNS — auto-create subdomains for each customer
- Brevo emails — welcome emails, payment reminders, expiry warnings
- Admin dashboard — customer list, server health, MRR tracking, audit logs
- Customer dashboard — server info, domain management, profile settings
One click on my admin panel and a customer has a fully running AI server. One missed payment and the system handles reminders, then cleanup. It's all automated.
The OAuth Problem
The hardest part — and I mean the part that kept me up at night — was authentication. Claude Code requires GitHub OAuth, and GitHub OAuth requires a redirect URI. On a local machine, that's localhost. On a remote VPS, there's no browser to redirect to.
The solution was a creative hack: the Jarvis dashboard on the customer's server initiates the OAuth flow, redirects to localhost (which fails), but the customer copies the failed URL back into the dashboard, which extracts the auth code and completes the exchange via SSH.
It's not pretty. But it works. And once you authenticate once, you never have to do it again.
One Person, One Product
I didn't hire a team. I didn't raise funding. I didn't spend months in product planning meetings. I had an AI agent, a server, and the skills I'd been building for months.
The entire Jarvis platform — control plane, payment system, provisioning, monitoring, customer dashboard, admin dashboard, landing page — was built by me and Tim. Two entities, one of them artificial. That's the team.
This is what running a one-person business with AI looks like in 2026. The barrier to entry isn't engineering talent anymore. It's knowing what to build and having the persistence to ship it.
Why Managed Servers, Not Just API Access?
I could have built a simpler product — a wrapper around Claude's API, a hosted chatbot, a prompt playground. But that's not what I use. And it's not what makes the AI agent powerful.
The power comes from server access. My AI agent doesn't just answer questions — it installs packages, writes deployment scripts, manages cron jobs, edits nginx configs, migrates entire servers. Take away the server, and you're left with a very expensive autocomplete.
That's the Jarvis thesis: give people a real server with a real AI agent. Not a sandbox. Not a playground. A production environment where the AI can actually do work.
Launch Day
Jarvis went live on April 7, 2026. Thai version first — that's my market, that's where I know the customers. The pricing starts at 990 THB/month (~$28) for the Spark tier and goes up to Omega for heavy workloads.
Every tier gets the same thing: a dedicated server, an AI agent, a chat interface, and a file browser. The tiers differ in CPU, RAM, and storage — but the experience is identical. Your AI agent has full root access to your server, period.
No lock-in. No proprietary platform. If you want to SSH in and do things manually, you can. If you want to install your own tools, go ahead. It's your server.
The product I wished existed when I started — an AI agent that actually does things, on a server I actually control — now exists. And if you've been reading this blog, watching me automate content, build personal tools, and run tasks in parallel, now you know: all of that is possible with Jarvis. Your own AI agent. Your own server. Ready in minutes.
— Pond