Sabitlenmiş Tweet
witcheer
14K posts
witcheer
@witcheer
community @NousResearch · sovereign compute advocate · ex @KPMG
Open Source Katılım Ekim 2021
1.1K Takip Edilen12.5K Takipçiler
It's worth watching the Hermes Agent Masterclass series Tonbi did -- really useful content!
tonbi@tonbistudio
Today's Hermes Agent Masterclass is the finale! Module 10 covers security, an important topic for running agents to ensure each agent can perform its tasks while minimizing exposure. Hermes has a ton of built-in security features. In this clip, I show how you can set approvals for each profile depending on your needs.
English
for all DGX Spark users, I found a very useful A to Z guide on NVIDIA that will let you have Hermes installed and connected to a local LLM served by vLLM (Qwen3.6-35B-A3B-NVFP4).
last updated June 12 so you might double check with Nous Research official documentation for latest tweaks
build.nvidia.com/spark/hermes-a…
English
For the past three months I've been tinkering with a self-hosted AI setup and figured I'd share the build and the lessons.
Mac Mini + Hermes + OpenCode, ~€30/mo.
Full read here pckt.blog/b/krzysu/my-20…
English
Hermes Wingtips #18: keep auto-learning, review the writes
if you want Hermes to keep learning without changing memory or skills unchecked, turn on both approval gates:
~ `/memory approval on`
~ `/skills approval on`
new writes wait for your yes instead of landing immediately.
English
please give us your recommendations!
emozilla@theemozilla
We're working on making the local model experience better in Hermes, what are the best local models at each weight class? My blind guess, please correct: 8-16 GB VRAM Gemma4 12B 24-32 GB VRAM Qwen3.6 27B Qwen3.6 35B 128 GB VRAM (Spark, M3 Max) ??? Can you do DSv4-Flash?
English
Opening the day with an image on Telegram created by Hermes Agent with everything running locally is just amazing!
English
night 2 of training the first speculative-decode draft head for Hermes-4.3-36B, on the one RTX 5090 that also serves it: the training context doubled to 2048.
that took gradient checkpointing inside EAGLE-3's TTT unroll. doubling the cap cost 6% in step time. max-length prices your token histogram, not the cap. most chat samples never reached 1024 in the first place.
fixed-config smoke at 26% of one epoch: code holds 1.29x over the bare teacher, and the workload that was below break-even crossed above it after the training started seeing full-length traces. trained-head precedent band: 1.7-2.2x.
field guide: github.com/notwitcheer/sm…
witcheer@witcheer
an EAGLE-3 draft head trains on the teacher's internal activations, not text. cache them to disk and a 36B teacher wants 4.5TB for the standard 54K-sample recipe. my disk is 570GB free only. so I chose online mode instead. the 4-bit teacher (19.5GiB, SGLang) stays resident and feeds the draft live, while fp32 optimizer state (8.4GiB) moves to system RAM with the step running on CPU. one more find: FSDP allocates gradient storage for frozen params, 1.5GiB for an embedding that never updates, so single-GPU skips the wrap entirely. ~~~ night 1 on one RTX 5090: 7,000 steps in 6h50m, expected acceptance 0.05 → 0.27 and still climbing. first drafter for Hermes-4.3-36B in training.
English
@DavidOndrej1 I haven't tried it in codex yet but using Hermes agent and even using the GPT-5.6-sol on high reasoning (not xhigh) is fucking amazing. Just don't ask it to do any UI stuff. It's not very good at UI design. I'd rather it suck at some things so that it could be better at others.
English
@catalinmpit even better now, go on Nous Portal, get Hermes on Cloud, connect to TG, done :)
English
It's official: for the first time ever I moved my Hermes agent to ChatGPT. That's how good 5.6 is
I was on Opus since day 1, even paying the API pricing (thousands a month), but I still thought the performance benefit was worth it
No longer. For the first time ever ChatGPT caught up.
Plus they are consumer friendly and let you use your subscription in the harness.
Now I'm getting better performance for a fraction of the price.
Never thought this day would come but it happened
Seen a lot of people asking the best way to do it. Here you go:
1. In your terminal type 'hermes dashboard'
2. This will pop open a dashboard in your browser
3. Go to 'Profiles'
4. In the default profile, click the 3 dots and click 'Change Model'
5. Make sure to choose codex/gpt-5.6-sol. Not the openrouter one if you have a chatgpt subscription
Boom you're good to go.
English
wake up babe, new marvelous feat just dropped ✨
Nous Research@NousResearch
GPT-5.6 is now supported in Hermes Agent and available via Nous Portal
English
Hermes Wingtips #17: check the built-in tools before you install a skill
a lot of popular third-party skills overlap with what Hermes ships by default.
five that people overlook: memory, web search, browser control, cron, and sub-agents.
all are native tools with no install :)
English
I can finally talk about this.
I’ve been using Hermes Agent for about 4 months now.
My life is considerably better than it was before using Hermes Agent. 🪽
English
I used Grok 4.5 through Hermes to automatically generate ~80% of the video below.
This is part of a series for Paid Subscribers on X and Exclusive Members on YouTube called Musk Industry Daily.
It autonomously does the following:
It finds the latest, most important news happening in the Musk Ecosystem from the last 24 hours.
Analyzes them against a knowledge repository that was cultivated from the corpus of my 1800+ videos, my 2 books, and my articles on X.
It writes a script that explains what's happening, why it's important, what people are saying about it, what the risks are, and what my specific take is.
It fact checks & sources the entire script and corrects any errors, plus improves any stylistic choices.
It generates slides that are used as b roll material for the video.
It runs it through the clone of my voice on ElevenLabs.
It packages all the materials up into a Google Drive, including the script, sourcing, cue cards for editors to know when to insert materials, the voice file, and a few other reference materials.
It runs daily at 12:30AM CT.
The editors then take all the materials, quickly package them together into the video, and upload it to my YouTube channel for exclusive members only.
Then an autonomous Hermes agent running Grok 4.5 will pick that file up, cross-post it to X for subscribers only, including a written summary of each item in the replies.
--
If you find this valuable, consider becoming a subscriber!
English
Grok 4.5 is now available in Hermes Agent!
Teknium 🪽@Teknium
Grok 4.5 is now available in Hermes Agent - You can access it through your Nous Portal subscription, Grok/X Subscriptions and API, and OpenRouter! Enjoy
English