Rubin, my AI assistant
It runs every day on my own server. It's where I find out what AI can really do before I suggest it to a client.
Why I built it
I worked across several tools at once: the browser, an AI chat, notes, a code editor. Every time I had to explain the context again. I wanted an assistant that already knows my projects, notes and open tasks, and answers without ten minutes of pasting.
I built it myself instead of buying a ready product because I wanted to understand how these systems behave in daily use: where they get things wrong, what they cost and how to keep them safe. That knowledge goes straight into the automations I build for businesses.
What it does every day
Code review
Finds patterns, summarises recent changes and points to the files behind a feature.
Notes and tasks
Searches my Obsidian notes, meeting summaries and open tasks.
Planning work
Writes a step-by-step plan that I then carry out or hand to another agent.
Web research
Searches, reads and summarises, so I get an answer in one paragraph instead of five tabs.
Reports and reminders
Sends me a daily summary and reminds me of tasks that have stalled for too long.
How it's built
One agent that runs all the time, connected to the tools and knowledge I use every day.
- ConversationA persistent terminal session I write to like a chat.
- AgentHermes agent, an open framework by Nous Research.
- ModelGPT-5.5 through my ChatGPT subscription, including image generation.
- ToolsTerminal, files, web, GitHub, memory, skills and scheduled jobs.
- KnowledgePersistent memory, my Obsidian notes and project repositories.
- ServerMy own VPS that's always on, not a chat tab I open and close.
Built on Hermes agent
I didn't write Hermes. It's an open agent framework by Nous Research, and I configured it around the way I work. Choosing its tools, memory and limits taught me more about agents than any course.
Safety
- Memory holds preferences and project context, never passwords, API keys or login details.
- Anything that can't be undone, like publishing code, deleting files or sending email, I approve first.
- Every tool has clear limits. Rubin doesn't run code outside the plan steps I've approved.
What this means for you
I test every idea on myself first. Only solutions that survive daily use reach your business, with the same rules: a person approves important steps, and your data stays under control.
Related project: PBI-MCP
An open MCP server that gives AI agents access to the Polish VAT white list and NBP exchange rates. Useful in automations that need to check a supplier or convert a currency.
Let's talk
Tell me which task takes most of your time. I will tell you whether it can be automated and what it costs.
Book a free call
Prefer to write or call?
Remote, anywhere in Poland. I reply within one working day.