An artificial intelligence assistant that runs my projects every day. It is not a proof of concept: it is the system I work with, and the reason I can speak from experience about agents with persistent knowledge of their own.
The challenge
Running several projects at once means constantly losing context: where things stand with each client, which domain expires, what task comes next, what was decided three weeks ago and why. Traditional tools store data, but they do not remember.
What I built
- Operations assistant accessible through Telegram that centralizes projects, tasks, clients, reminders and context. You write to it like a person.
- Persistent memory across sessions, with search over the full history. I never have to repeat anything.
- Scheduled automations for periodic website checks, recording reminders, domain expirations and client follow-up.
- Obsidian as the local control center and Notion for projects that require collaboration.
- A Mac running continuously using
pmset, LaunchAgents andcaffeinate, with private access between machines through Tailscale and SSH. - Multi-agent workflows with separate roles for product, development, user experience, security and artificial intelligence.
- Real production deployments with a verified backup beforehand and route-by-route checks afterward.
The result
The system monitors the sites I manage, alerts me when something breaks, maintains the context of each project and deploys changes to production with a backup first.
This project is why I can deploy AI agents with practical judgment. When someone asks whether an AI agent would work for their business, I do not answer from theory: I answer from the agent I use every day, including its limits and failures.
See how I apply this to other businesses →
Want something like this for your business? Book an initial 30-minute consultation or message me on WhatsApp at (+1) 407 616-7092.
