No pricey Mac Studios and no budget-burning API bills: how I combined Hermes, Google Antigravity, and a Gemini Ultra subscription on a tiny server at the lowest possible cost.
If you’re a solo founder building projects on your own, you already know the bottleneck today is no longer writing code or executing tasks.
AI tools made getting things done ten times faster, yet they multiplied the mental chaos and scattered ideas tenfold too.
For years, my problem wasn’t how to build, it was how not to lose the thread.
Midweek, say, I’ll hit a wave of excitement and deep focus, start a new project, and open forty browser tabs. Two days later, a second idea jumps into my head. By the time I come back to the first project the following week, it feels like someone else started it: Where did I leave off? What’s left before launch? And why did I drop it in the first place?
When I looked for a way to run an autonomous AI agent 24/7 as a partner to hold the thread and organize the mess, I ran into a weird wave of flexing: people buying thousand-dollar Mac Minis and Mac Studios, or hooking their agents up to API keys that burn $300 to $500 a month in endless loops.
As for me, I’m building from my "cave" and documenting my run from a real operating budget of just $900 to my first $1,000,000. Burning hundreds of dollars a month on bills means running out of fuel before I even get started.
So I went with a very practical, different setup: I paired the open-source Hermes Agent running on a modest little cloud VPS (just 4GB RAM) with Google Antigravity backed by a Gemini Ultra subscription.
This system is running alongside me right now as you read these lines, and here is how I actually use it every single day:
1. Turning Telegram Into a Live Operations Room (Text & Voice)
If I built a custom web dashboard, I’d get excited about it for two days and then forget to open it forever.
Instead, I connected the agent to a private, locked-down Telegram group split into clean workspaces: one space for server monitoring, one for building projects, and one for publishing and marketing.
While walking down the street, I send a voice note rambling through whatever is on my mind. Within seconds, the system transcribes it, understands what needs to be done, applies the changes or drafts the plan, and replies with the finished result.
2. Running Google Antigravity as an External Brain Without Choking the Server
This tiny server isn’t sitting idle. At the same time, it runs several live websites, projects, and automation tools.
If I tried running heavy AI models directly on it, it would crash immediately. So I made Hermes a lightweight coordinator quietly managing tasks in the background, while all the heavy thinking, coding, and image generation or editing happen by linking it to Google Antigravity and a Gemini Ultra subscription with maximum usage limits.
3. Persistent Memory That Stops Context From Slipping Away
Standard AI chats forget what you agreed on yesterday as soon as the session gets long. Here, whenever we settle on a specific workflow, a standing rule, or publishing steps, the agent saves it right away as a permanent skill and loads it automatically on every future task.
On top of that, the agent maintains a private, isolated documentation archive for my projects and notes, linked to a pinned message inside the control room. One tap lets me review the status of every project and post without drowning in old chat history.
4. Building and Automating Entire Projects With My Final Sign-Off
Through this agent, I code full applications and build custom software solutions for my projects from scratch, even when I’m nowhere near my laptop. I explain an idea for a new tool or a feature I want in an existing project, and it writes the code, runs it to test everything itself, and fixes any errors that pop up until the build is ready to go, right alongside monitoring my projects 24/7, designing visuals, and preparing content on a single command. Even with all that power, it operates under one strict rule: it never touches a sensitive part of a live project or publishes a single word publicly before showing me the result on Telegram and waiting for me to hit "Approve."
Now for the biggest question: how do I get a subscription that normally costs $150 to $300 a month for just $30?
That is the best part of this whole setup.
I don’t pay that steep retail price, and I don’t use shady workarounds that get accounts banned. Instead, I use a smart setup that gives me an official Google AI Ultra subscription with full usage limits inside Google Antigravity for around $30 a month.
And since I decided from day one never to sell courses, templates, or "paid secrets," I put the full breakdown of this method, complete with the practical steps and the exact providers I personally use, in a free post waiting for you inside my private Telegram channel.
In there, inside the "cave," I document with total honesty every project I build, every experiment that works or flops, and every single dollar I spend or earn starting from $900 toward my big target.
Join the channel here to read the full method as soon as you step inside, and follow along on the road to the first $1,000,000... step by step, zero filters:
The Dark AI Engineer: Road to 1M 🦇