Hermes Agent

Does the desktop app still install a full Hermes Agent setup or is it client only by now? I have Hermes running on a different server. I wanted to install the desktop app as a client interface for that server setup but it wanted to install Hermes on my MacBook. I know I can point the desktop app to my server install later but I don’t want or need to have Hermes installed on my MacBook just to use the desktop app as a client.

Is there a way you keep the gateway running. I noticed that not the hermes gateway btu the photon gateway or something ended (thats what hermes said). the hermes gateway start was on but not the connection to photon or something like that

Only Claude, other models work fine (OpenAI and opencode)

That is the problem, it didn't have root access anyehere, it took my ssh key from my user acc it is running on a vps somewhere in german OVH DC

Also, Hermes guardrails / approval tool call relies on regexes, and changing ansible config or similar isn't just within the cases those catch.

The scores could use some sharpening up for sure

Dont blame Hermes for what could be the fault of the AI model

I like the multi thread experience of grok bot, it's slick without context rot like you get when trying to recreate that in telegram for your Hermes agent.

If you could get me a good mobile app that will let me seamlessly continue my work on the move, I will cancel all my other subscriptions and go for Nous.

After spending some time learning to use hermes properly, I'm really impressed. It's almost perfect.

The process that works for me right now is leveraging Claude code to get a Hermes capability and any supporting process(es) to production, and then Hermes with deepseek or glm to execute day-to-day, learn, and marginal improvements. I still rely on Claude for bigger functional leaps, though. It works for me and seems to get me to a working solution faster than just Hermes solo.

I have mine note problems in a to-do list so he can research them later in a cron job. That was the entirety of my suggestion. Here is the last update I got (the table didn't cut and paste well from telegram), and this is just the flash model. It's basically become a running system health and the small stuff gets fixed automatically. He leaves the bug stuff for me, which is usually "file a bug for that so Neo can fix it". I could never keep up with all this without him!

I love hermes because it's already in the channels I use (Whatsapp). There are absolutely problems with this approach but i find myself defaulting to voice (openAI) or chat (Hermes on Whatsapp) for my on the go-mobile AI use cases.

2-6 minutes on average => it has pretty heavy model context. this is what happens when i say hello. lol. you gotta strip off those unneeded guts one by one.

The Desktop can render it and it looks alot better than whatever that notation is

Seeing that Nous promotes plugins system I tried to develop a simple module for that. According to my Hermes however, there is currently no proper hook in Hermes code. It could develop smart model routing within the same provider allegedly, but there is no way to cover multiple providers without source code change and PR.

But hermes used to let me tell it to reboot. No logical reason for me to not to be able to toggle this saftey feature off.

To not be able to tell it to restart is kind of a key feature missing.

Heh, yeah out of the box Hermes can be a handful. We made a virtual "stone". As in, chisel this data into stone, it is not changing. We populate our wiki with data from the stone. It kinda works, mostly.

the agent talks to Matrix through normal messages only. there are no Matrix-specific tools for creating rooms, inviting people or redacting.

end-to-end encryption is opt-in. you install libolm plus the Matrix extras and set MATRIX_E2EE_MODE=required, which fails closed rather than falling back to unencrypted.

lock it down with both MATRIX_ALLOWED_USERS and MATRIX_ALLOWED_ROOMS. with either unset, anyone who reaches the bot in a joined room, or any room it joins, can start a turn.

I built an AI system using Hermes and that thing can basically build anything at this point

I wouldn't even worry about some of that stuff like the phone app.

Hermes will probably be more for the technical inclined

I kept my Hermes default profile for quick chat or small changes in my homelab as it's more responsive and doesn't need to go through the hole hierarchy.

My #1 reason for building first on Hermes. I do software used across the world and the fact the US can turn off access to your AI with little notice is a big concern of customers. I’m glad to support codex and others but I wouldn’t want to walk into a meeting without an OS option.

hermes agent needs at least 64K context for tool work, and computer use eats context fastest: each screenshot counts as ~1,500 tokens, and a 20-action session uses ~30K on screenshots alone. at 32k your original request gets pushed out of the window partway through the run, so it forgot your site and made up amazon.com.

I run Hermes because I can have exact and precise control over certain tasks. I can have it load a page and refresh it every second for hours if I want, waiting for action. It’s just more powerful if you’re willing to deal with being your own maintainer.

hermes appears to be pretty good at finding what I need in sessions.