hermes delivers!
Posted by Emotional-Breath-838@reddit | LocalLLaMA | View on Reddit | 8 comments
running: Qwen3.5-9B on Mac Mini 24GB and Hermes Agent via WhatsApp.
step 1. tell Hermes to create a skill called X.com. the skill must allow me to paste X posts to WhatsApp (Hermes has its own phone number via WhatsApp for Business) and review what i sent. then, provide me with three choices: find the repo and build it, understand it (and rememeber it) or other.
step 2. stop bookmarking things on X. just hit share and drop it on Hermes. Hermes will eventually send you a whatsapp
message that its done
step 3. let people on Reddit know that we live in a post-OpenClaw world and its getting better, faster
in the example screenshot, someone on X was bragging about their stock portfolio management software. built in AI, up to date quotes, algorithm trading, etc. so, i just dropped it into Hermes' whatsapp and said build this same thing but i dont want to pay any api fees so figure it out.
hermes allows me to spin up additional sub-agents as needed so ill eventually have one that does trading for me on a limited budget.
Substantial__Unit@reddit
I have mine running but its using Claude. What benefits would I see, besides privacy, running a local llm
Ok-Internal9317@reddit
Yes, Hermes is by far the most promising, opencode isn't far off in terms of single message completion, but it need constant prompting. Only problem with hermes is that its a bit costly to run 24/7 and doesn't have provider discounts like opencode go
Emotional-Breath-838@reddit (OP)
hermes runs off a local so its free.
yes2matt@reddit
be cool if our agents had a by-agent-for-agent discord or subreddit that they could post skills/tools and look up others. It would not need to be human-readable exactly.
Distinct-Shoulder592@reddit
I think of LLM systems like a loop that keeps getting smarter. A Hermes-style agent handles quick decisions in the moment, while an LLM Wiki Compiler turns those actions into organized knowledge that keeps building and improving over time.
sidesw1pe@reddit
Is there a way Hermes can access your X account and monitor your bookmarks? Then you can bookmark something and it could regularly take your updated bookmarks and perform your workflow, and remove the bookmark while it’s at it.
Emotional-Breath-838@reddit (OP)
Yes. You would need to sign up for the C api. Tell Claude code to generate a Hermes skill that does exactly what you want and ensure it knows about the api, your expectations and the memory space you want it to use.
Big-Ad1693@reddit
Ich bin auch total begeistert, hab das 35ba3b model mit 128k Kontext auf meiner p40
Läuft mit 50t/s ca und normal ist das nicht soooo smart das Modell, ich bevorzuge das 27B aber 15 bs 50 t/s merkt man schon, auf jedenfall kommt das Model und Hermes Agent super miteinander aus, bin absolut begeistert
Gestern bis halb 4 zugeschaut wie es arbeitet, Nacht vorbei 🤣