My Harness Setup With Hermes
I’ve been experimenting with personal AI agents, and Hermes has gradually become part of my daily routine. I use Hindsight for long term memory, helping Hermes retain useful context across sessions. I mainly interact with it through Telegram, where I receive updates, learning material, and results from scheduled tasks.
My Setup
I run Hermes on a remote VPS with 4 CPU cores and 8 GB RAM. This allows it to run continuously without depending on my personal machine.
Telegram is my main interface for interacting with Hermes and receiving its outputs. I’ve also created a dedicated GitHub account for it, giving it access to selected repositories instead of my personal account.
What I Use It For
- AI daily digest: Every day, Hermes sends me an AI generated digest that helps me keep up with useful information without having to check multiple sources manually.

- Server monitoring: I have scheduled cron jobs that check server health, along with a separate watcher that keeps an eye on disk usage and alerts me when something needs attention.

- Daily Vocabulary: Every day, Hermes sends me three English words with their meanings, pronunciation, synonyms, and example sentences. That is over 1,000 words a year, building a useful vocabulary over time.

- Qur’an Lessons: Hermes reads The Clear Quran by Mustafa Khattab from a PDF using Python’s MuPDF package. It extracts the verses and prepares a daily lesson in a thematic format, keeping the content manageable and avoiding unnecessary repetition.

- Islamic Fatwa Watcher: Hermes monitors an online Islamic fatwa website originally published in Urdu. It collects rulings, translates and summarizes them in English, and includes a link to the original source for reference.

- Telegram and websites: I use Hermes to host and orchestrate other Telegram bots.

- Hermes backups: I have a backup routine for my Hermes setup itself, keeping one copy on the machine and another in Telegram so I have an additional copy available if something goes wrong.

- Infrastructure: I self host Umami Analytics and a few other apps using Coolify, a self hosted alternative to platforms like Vercel and Railway. When needed, I ask Hermes to check things or help manage deployments.
What’s Next
I want to improve long term memory, make scheduled workflows more reliable, and connect my existing tools in a better way. The focus is on making the current setup more useful rather than adding automation just for the sake of it.
Final Thoughts
The setup is still evolving, and there’s plenty left to improve. For now, I’m more interested in making the workflows I already have reliable than in constantly adding new ones.