I know a few of you have run Hermes Workspace.
Have any of you self-hosted it successfully?
I've been banging my head against it all day. For something that should "just work," it sure isn't. The gateway isn't finding the agent install. The web UI isn't connecting to the back end (and I don't even know what that's referring to). But, agent is installed because I can SSH into the server and run the agent easy peasy.
I don't know what else to do at this point.
#Hermes #Hermesagent #hermesworkspace #selfhosted #agents #asknostr
Login to reply
Replies (25)
honestly sounds like the docs are written for people who already know the architecture, not for folks actually trying to deploy it cold.
ποΈ Documenting our family's daily reality in Gaza on my pinned post if you'd like to check it out.
Hit backend with port 9119 in the UI
@The Beave the agent working over ssh but not through the gateway usually means the gateway runs as a different user with a different path, so it never finds the install you can see. i would check the service user first, then whether the back end is listening on 127.0.0.1 rather than 0.0.0.0.
Just running the agent sorry.
This might sound dumb... But, how?
And, also, I don't understand what exactly the "back end" is in this context.
Yeah. That's been a huge part of this. I've edited the .env file a bunch and thought I'd set it to 0.0.0.0, but still no dice.
The agent by itself is pretty darn good.
@The Beave check what it is actually bound to with ss -tlnp before you trust the .env, since a lot of these apps only read that file at startup and some ignore the key entirely. if any of it is in docker, 0.0.0.0 inside the container does nothing until the port is published, and the gateway has to reach the backend by container name instead of localhost.
It's running in a Debian LXC on my proxmox host.
It sure doesn't seem to be paying attention to changes in the env file, but... I'm suspecting that there's another env file that I'm not aware of that is actually being read.
The frustrating thing is that one not heard a peep of install issues from anyone else, even poking around for days. Being ahead of curve sucks sometimes. I'm not knowledgeable enough to be doing any of this as a pioneer. π€£
@The Beave in an lxc there is no port to publish, so the systemd unit is a likelier culprit than a second .env. run systemctl cat on the unit and look for EnvironmentFile or inline Environment lines, because those override anything sitting in the app directory. systemctl show <unit> -p Environment tells you what the process was actually handed, and cat /proc/<pid>/environ confirms it on the running process.
I honestly don't know how to do that. I don't know what you are referring to with "unit." π
I like it but pi.dev is my favourite :)
@The Beave a unit is just the text file systemd uses to start your app, usually sitting in /etc/systemd/system/. run systemctl list-units -t service | grep -i hermes to get its name, then systemctl cat that name to print the file. if there is no EnvironmentFile line pointing at your .env, systemd never reads that file at all, and that is why your edits do nothing.
I haven't messed with that. π
Ah! I see. I'll poke at that after work. Thanks.
Been using it for months now. Works great.
I'm assuming that you pay for model use and aren't self-hosting everything. Correct?
Hermes + Ollama. Thatβs what I use.
But yes, I run Ollama cloud, though really that works just as you would a local LLM. Unless you mean local agents are not smart enough to run Hermes.
I think that's my issue is I'm running everything self-hosted.
Finding and tweaking the right model has made a huge difference to getting mostly useful output from my own hardware.
I just tested a full self-hosted setup with Ollama.
The only extra step that I needed to do is run
Ollama serve <model name>
Then Hermes quickly detected the model over http://localhost:11434/ or maybe it was
Http://localhost:11434/v1 I forgot which one worked.
After that, it ran as expected.
I have no issues with Hermes agent.
I can't get workspace to talk to the gateway which connects to agent. Plus it doesn't want to connect to Ollama, but I think I fixed that.
I see. Not sure where to start.
There is a default profile that you start with.
Maybe try > Hermes doctor ?
Here is why my agent says, hope it helps π€·ββοΈ
The main diagnostic command is:
hermes doctor
It checks dependencies, config, providers, connectivity, and flags issues. Your friend should also run:
hermes doctor --fix
That auto-repairs common problems.
For the gateway connection issue specifically:
hermes gateway status
hermes gateway restart
And to check what's actually wrong:
hermes status --all
Quick full reset if gateway is stuck:
hermes doctor --fix
hermes gateway restart
Ah! I hadn't seen or read the "Hermes status --all" command before. I'll run that.
The rest of it has come up in various trial runs of getting this working.
I currently have figured it that the Hermes agent that the workspace installer installed can actually muck about with the system it is installed on, so I told it to go fix itself.
Probably won't work, but why not try? π€£π€£π€£
Go fix yourself should be the slogan for Agentic AI.
So... I told it to fix the issue and keep going until it can report that it is working. Surprisingly, it is doing the steps that it's big brother (gpt-oss, I'm running gpt-oss: 20b on my system, currently, as it works very well so far with my hardware limitations) suggested and now I don't need to copy and paste, so... This might work? Maybe? We'll see.