My frustrating experience with Hermes
I am not a dev, I don’t necessarily like CLIs, but don’t feel intimidated by them either. I’m fairly familiar to technology, including automation and LLMs in general. I’ve setup N8N/make automations and connected to front end (vibe coded) a few times and use ChatGPT, Claude/Claude Code, Gemini and perplexity heavily every day.
Never tried OpenClaw or anything of the sort. Hermes is my first experience.
The hype is:
Run one command, setup your provider, setup Telegram if you wish, off to the races, the system should improve itself the more you use it. In one month it’s running your life and making you money while you sleep.
My expectation:
Some fiddling necessary, rough ride on the first 10-20 hours to get the hang of it, maybe one or two reinstalls to start with a clean slate, first week is a steep learning curve, second week it starts to smooth out and one or two light but helpful cron jobs would start to show the ROI on the horizon. By week 3 I thought bugs would be present but somewhat easy to sort out, especially as I don’t mess with the system heavily and try to keep it healthy by only doing clean installs, not fiddling with things I don’t understand, etc.
If I could make it a little more autonomous than Claude Cowork, focused on boring admin tasks to help me be a bit more productive and organised, I’d be happy for now, until I learn where I can push it to help me with other stuff.
Reality:
3 reinstalls in 4 days to begin with. Gateway problems, sometimes it would work, sometimes wouldn’t. I downloaded Termius to help but it’s not great on mobile with long texts because it used to break once you tried to zoom/scroll. I think this was a Termius bug because it doesn’t happen as much anymore.
Fine. First few hours. Expected.
I then decided to install it bare metal. Gateway issues magically gone. This was the fourth install on the first week. With telegram working more reliably I could actually have a proper chat with it and have continuity from my Mac to my iPhone while on the go. It’s underrated how much this helps. I now get where the hype comes from.
Connected Notion, fine. Started building the system and documenting. For an agent that should have a better memory system, it’s frustrating to see that it “forgets” instructions from one day to the next. ChatGPT and Claude are miles sharper. Fine, let’s resolve the memory issue with external persistent memory. Setup Open Brain, then didn’t think this was a good long term option, changed to Hindsight. Apparently working fine.
Took me 2 days to get the Google workspace connected. Once I did, it worked for 2 days before silently breaking because Hermes tried to “improve” it and accepted a bearer with less permissions than it had before. The plugin was working perfectly and Hermes admitted to breaking it while doing something completely out of scope. I only found out when I needed to use it.
TTS also broke silently, and I find out when trying to get something done in the little free time I have in my day.
Setup Firecrawl to test some research capabilities and the result was ~20% of what I would have gotten with my Claude skills if I had applied the same time. Depressing, unusable. A few more rounds and I gave up.
As I used it through week 2, I started to get annoyed that it wouldn’t follow my detections very well (fine, maybe I’m doing something wrong), or I would burn the time I had set aside to make something productive with it with troubleshooting or settings that wouldn’t work as it described.
I noticed that instead of saving my time it was draining it.
It’s now week 3. I’ve spent about 50 hours altogether from my first ever contact with Hermes.
I’m reducing the time I’m investing in this now. It does not pay off. At least not for me, not by now.
Once again I open Telegram to setup a cron job with a research and reminders and I’m greeted with a cron report about a broken script. Once again I have to read long texts at midnight to try to understand what is happening under the hood and try not to make things worse. Once again the little time I had was replaced by troubleshooting. Once again I think that I would have more time if I just ditch Hermes for now, until it’s out of beta.
I really love tinkering with tech, and I’d probably have it setup just to play with it on my free time. But free time is a very scarce resource for me right now, and I was hoping I would at least break even by week 3 after the initial learning curve and setup was done. But all I see is that the early-adoption tax is not a joke with Hermes (and other similar agents).
My sentiment is that I can’t rely on it for anything that isn’t being closely monitored in semi-real time or anything that requires a minimum level of reliability.
Opening Hermes now feels more like “let’s see how I’m gonna waste the next hour troubleshooting something that was working perfectly last time” than actually “let’s try this idea”.
Sorry for the rant, thanks for reading it.
Just wanted to know if I am so stupid that I can’t make it work somewhat reliably or seamlessly (if it’s at all possible with beta products), or if this is the average ride and I just fell for the hype?