This is still mostly solution in a search for a problem but very satisfying side vibe coded thing. last tools and cron jobs are closest so far to feeling more connected to what Hermes is doing in background
Added few things so it's more usable overall: moved labels and wifi settings to separate file. Source at https://github.com/rudidev08/tab5-lcars It's mostly fidget toy for me, but very satisfying at that! (grew up with star trek).
Ways to use it:
- run it on device with preset tab and static text, so no network connection needed
- it can refresh data from local server
- there's voice mode i've used with hermes + hermes webui too
This was quick vibe coded experiment while i was (unsuccessfully) trying to make UI i liked, and it should be pretty easy to ask LLM to make changes.
Supports sounds but you'll need to run a script to download LCARS sounds.
P.S yes I got tab5 before i knew what i wanted to use it for :)
P.P.S. had keyboard support in beginning but felt very clunky for UI and i forgot about it, so i don't know if it still works correctly or not. voice mode was accurate enough not to reach for keyboard.
built with kimi. I’m both excited and not thrilled because I’m spending lots of time working on custom UI, and this was a 5 min experiment, and it looks way better than my stuff. right now it’s mostly a good fidget toy because of LCARS sounds. connects with local Hermes agent
Posted source: https://github.com/rudidev08/tab5-lcars added option to show static text (no network connection needed) or any data from basic server.
for nous research subscription:
- do I still keep rollover credits if plan switches to free?
- can bought credits use paid models on free plan? (or is subscription required to use all features)
There are few hermes tweaks and bug fixes that are quick one liner for me.
Is best way to maintain this is to fork hermes repo, and update from there? I usually commit github PRs and patches, so i'm not sure what's most effective way to keep set of changes on top of official changes without breaking anything.
(for example, approval text is one line on desktop but many on gateway, and i'd like to patch that to see diff before approve/ask/deny)
concurrency setting helped me with random crashes, but now I sometimes have STT and small aux models waiting too long. Is there a way to have separate queue settings for them?
or, maybe run another oMLX or such to serve aux and speech only?
https://github.com/NousResearch/hermes-agent/pull/58698
This is huge for me, since manual approval only gates specific commands. Plugin based means you choose which tools get the prompt, and then hermes keeps a list of what's allowed, asked, blocked.
Currently, i'm going to try this with asking me for anything, and see how it goes.
Now with this, i can fully move on to Hermes (until now i had to juggle Goose and n8n for more sensitive operations)
Edit: might have spoken too soon, Claude mentioned it might only work for CLI and not desktop. Testing now and will open an issue if so
Edit2: yea confirmed, desktop won't show approval buttons (but it does hit the plugin!) Off to write an issue for it
Edit3: no it's reverted :( https://github.com/NousResearch/hermes-agent/commit/74cc9ee3f06ab09d7f6b06103b961722a2153578
As far as i can tell, there's currently no way to do granular approval for individual changes. Ideally i'd be able to see a diff of proposed changes i can approve or deny (file edits, file move, etc). I did some research but I like that native solution would be supported in chat and gateway. It's personal assistant work for low number of docs with high importance.
My current workaround i'm testing is having hermes work in docker with copy of my files, and then i see rsync diff with changes before applying them to main one.
How do y'all do this?
Edit: github mention gave me idea: i have carbon copy cloner for backup, i could pick folders hermes works with and backup those. need to see how good the diff is.
Building first local LLM setup. if i'm using Open WebUI with Hermes Agent, should i turn off memory from Open WebUI for this specific case? since Hermes agent has it's own memory and improvement loop?