Exploring Hermes

I personally have a principle that everything is worth trying, in that vein I went on a weekend exploration adding Hermes to my home server. My home-server is a small refurbished minisforum, nothing special but I enjoy it so much to do tinkering. I started with a few goals that I wanted to get out of Hermes before starting this little weekend project

I have an Obsidian vault where I store all of my work that I do for OSS, my daily journaling, yearly/monthly reflections, thoughts, … There’s a lot in there but I have issues with being consistent with writing in my journal and doing my reflections. I would also often forget to look back at analysis that I have done before, believe me I have investigated the same Preact issue more times than I’d like to admit.

I used factory default Hermes, put it on the box and connected Telegram to it. I asked it to set up a few cron-jobs with appropriate skills to perform these investigations and finally I added Tavily to give Hermes access to internet search, … I am not too fancy to do different models and all of that fun stuff so I set it up to just use GPT5.4 using my OpenAI subscription.

One thing that crossed my mind a lot is that I’d love to have this run fully locally, where I can do my own inference but looking over the system requirements for that I quickly backed out… My dev-box does not have a GPU and my gaming PC does not have a GPU that’s on (or available) 24/7 to do this compute. I am hoping that TPU’s come out quickly and are accessible to consumers so we can do this tinkering.

Now, all of this exploration brought me to a point where each Saturday I get research delivered to my inbox about my open-source projects and it has actually been actionable. It surfaced a repository where I forgot to add trusted publishing and it has actually convinced me to use Codex security scans on my most security critical repositories. Every morning it provides me with a news digest of things I care about, that’s mainly economic news in Europe and developments in tech.

Having been critical about flows like OpenClaw and Hermes, I am actually converting a bit to being a big fan of it. During dog-walks and just general boring moments I can just send ideas that pop in my head to my agent over Telegram and be sure that it won’t get lost. The agent has access to my synced Obsidian vault and will just persist my thought which I can work on when I get home, when I feel adventurous I can even ask it to do preliminary prototyping or research.

I need to find ways to simplify my maintenance burden as my life is changing a lot in the coming year, offloading tasks to Hermes that I can define well is going to be a superpower which I hope to harness well by the time change arrives.

One of the things I’d love to do in the future is give it access to the Twitter API so it can watch topics like Preact and notify me about people talking about it. I’d love to see more of the projects people create with tech I work on!

I am new to all of this and I am very open to anyone telling me how to do better or share their experience with me, I’d honestly love to hear from y’all!

In our new white paper, The Architecture Behind AI-Native Revenue Automation, Tabs CTO Deepak Bapat breaks down what it actually takes to apply AI to revenue workflows without breaking the books.

You’ll learn why probabilistic reasoning isn’t enough for finance, how Tabs pairs LLMs with deterministic logic, and why a unified Commercial Graph is the foundation for scalable, audit-ready automation. From contract interpretation to cash application, this paper goes deep on where AI belongs—and where it absolutely doesn’t.

If you’re evaluating AI for billing, collections, or revenue operations, this is the architecture perspective most vendors won’t show you.

Get the whitepaper