The Nemotron 3.5 Bolt Hermes combination gives a business a private AI agent that runs on its own hardware, using NVIDIA's new Nemotron 3.5 Bolt model as the brain of Hermes Agent.
Bolt is NVIDIA's 74B-parameter vision-language model in a native 2-bit build, designed to fit in 24GB of memory and run everyday agent tasks locally.
Hermes Agent, the free open-source agent from Nous Research, added Bolt to its one-click local runtime in v0.22.3 on 21 October 2026.
NVIDIA's model card says it's ready for commercial or non-commercial use under the OpenMDW-1.1 licence, which matters if you plan to use it for client work.
In this post I'll explain what it is, what hardware you need, how to set it up, and six practical ways a business can use it, with the limits you need to know first.
If you want my team to look at where AI agents and AI search fit your business, book a free SEO strategy session with my team.
Nemotron 3.5 Bolt Hermes In One Table
Here's the whole thing on one screen, checked on NVIDIA's own pages and the official Hermes docs and release notes.
| Question | Answer |
|---|---|
| What is it? | It is NVIDIA's local vision-language model running as the brain of Hermes Agent. |
| Who made it? | NVIDIA made the model, and Nous Research makes Hermes Agent. |
| How big is it? | It has 74B total parameters and 9B active parameters. |
| How is it squeezed down? | NVIDIA uses native 2-bit quantisation with quantisation-aware training to fit it in 24GB. |
| What can it read? | It reads text, images and short videos, and it writes text. |
| What does it cost? | The model and Hermes are free to download, and you pay for the hardware and power to run it. |
| Can I use it for business? | NVIDIA's card says it's ready for commercial or non-commercial use under OpenMDW-1.1. |
| When did it land? | NVIDIA released it on 15 October 2026, and Hermes added it on 21 October 2026. |
🔥 Want my exact Nemotron 3.5 Bolt Hermes setup, step by step? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
What Nemotron 3.5 Bolt Is, In Plain English
Think of Bolt as a big, clever model that NVIDIA has shrunk to fit on one serious workstation.
NVIDIA's model card says it was compressed from the bigger 120B-parameter Nemotron 3.5 Super VL model, then quantised to 2 bits so it fits in 24GB of memory.
The card says it's built to run everyday agentic tasks locally and run alongside your other applications.
It's optimised for local agent harnesses and tasks involving tool calling, web browsing and terminal use.
It can also look at images, which means screenshots, charts and photos of documents are fair game.
Hermes Agent is the harness that puts it to work, with memory, skills, tools and a messaging gateway around the model.
The Problem It Solves For A Business
Here's the before state I see in a lot of businesses.
They love what AI agents can do, but they're nervous about sending client files to a cloud model, or they're worried about the bill growing every month.
The Hermes Local Models docs say local models run entirely on your own machine, with no account, no API key and no network access after the model is downloaded.
On my own GoldieBench board, the small local models score far below the big cloud ones, which matches what most people feel when they try them.
NVIDIA's blog says its tests show that compressing a large model to fit 24GB beats training a smaller one from scratch for the same size limit.
That's the pitch in one line, because you get a bigger brain in the same memory.
What Hardware Your Business Needs
This is the honest part, because Bolt isn't for every laptop.
NVIDIA's GGUF model card gives two setups for running it with llama.cpp.
| Hardware | What NVIDIA's recipe gives you |
|---|---|
| RTX Spark 128GB or an RTX card with 48GB+ | You get a 262,144-token context and up to 7 draft tokens for speed. |
| RTX Spark 64GB or an RTX card with 32GB+ | You get a 65,536-token context, up to 6 draft tokens and a compressed KV cache. |
The card lists NVIDIA Ada Lovelace and Blackwell GPUs, running Windows or Linux.
NVIDIA doesn't mention Macs on the card, so I can't tell you how it runs on Apple hardware.
If your team doesn't have a machine like that, the bigger Super VL model through NVIDIA NIM is the cloud alternative, which I cover below.
How To Set It Up In Hermes
You don't need to touch the model files yourself.
- Update Hermes to v0.22.3 or newer, with
hermes updateon the command line or through the Store if you use the Windows Store app. - Open Settings, then Providers, then Local Models, and click Install runtime.
- Find Nemotron 3.5 Bolt in the catalog and click Download.
- Check the memory fit colour, then click Use.
The docs say green means it fits your GPU, amber means it spills into system RAM and runs slower, and red means it's too big for the machine.
In a terminal, the docs say hermes model then Local models does the same job.
Plan for roughly 25 GB of downloads, because the Hermes catalog pulls the model, a vision file and a small draft model for speed.
🔥 Want the exact agent setup I run my business on? Inside the AI Profit Boardroom, I've got a full Hermes and local models section with step-by-step tutorials, plus the installable Agent OS. You'll be learning with 3,400+ members and four coaching calls every week. → Get access here
6 Ways A Business Can Use Nemotron 3.5 Bolt In Hermes
These uses come from what NVIDIA says the model is built for and what Hermes documents, not from income promises.
1. Working on private client files
Agencies and consultants handle documents clients don't want in the cloud.
With Bolt running in Hermes Local Models, the docs say nothing leaves your computer after the download.
That lets you summarise, sort and draft from client files on your own machine.
Still check your client contracts, because privacy rules are about more than where the model runs.
2. Reading screenshots, charts and scanned pages
NVIDIA's card says Bolt takes RGB images in JPEG or PNG, plus MP4 videos up to two minutes.
A business can hand it screenshots of dashboards, charts from reports or photos of paper forms and ask for a written summary.
I'd use this for turning a messy analytics screenshot into three bullet points for a client update.
3. Web research that runs on your own machine
NVIDIA says Bolt is built for web browsing, and its own benchmark table shows it scoring 66.75 on BrowseComp.
Inside Hermes, that means research jobs like competitor checks or keyword ideas can run on a local brain.
Always ask the agent for sources and check them, because no model gets research right every time.
4. Terminal and admin tasks
NVIDIA lists terminal use as a target task, and its table shows Bolt scoring 50.98 on Terminal-Bench 2.1.
Hermes gives it the terminal, with the docs saying dangerous commands need approval before they run.
A small business can use that for file tidying, report generation and simple scripts without paying per message.
5. A local brain for your always-on agents
Hermes can run messaging bots and scheduled jobs, and a local brain means those jobs don't add to a cloud bill.
The Hermes docs say idle local models are unloaded after 15 minutes to free GPU memory and reload on the next message.
That suits a workstation that sits in the office and answers when your team needs it.
6. A cloud fallback for the heavy jobs
Some jobs need a bigger brain than one workstation can run.
The Hermes docs show how to use Nemotron 3.5 Super VL through the NVIDIA NIM provider, which NVIDIA lists at 120B total and 12B active parameters with up to 1M tokens of context.
That gives you the same model family for small jobs on your own machine and big jobs in the cloud.
If you want help working out which of these fits your business first, grab a free strategy session with my team and we'll look at your site, your offers and where agents can save you the most time.
🚀 Ready to put Nemotron 3.5 Bolt Hermes to work in your business? Inside the AI Profit Boardroom you get step-by-step video trainings, my installable Agent OS and four coaching calls a week with 3,400+ members building real automations. → Join the AI Profit Boardroom
How To Add Super VL For Bigger Jobs
The Hermes docs give this setup for NVIDIA's hosted service.
First, you create a free API key at build.nvidia.com and add it to ~/.hermes/.env.
NVIDIA_API_KEY=nvapi-...
Then you start a chat with this command.
hermes chat --provider nvidia --model nvidia/nemotron-3.5-super-vl-120b-a12b
The docs say a local NIM on a DGX Spark, a DGX Station or a local GPU works too, by setting NVIDIA_BASE_URL to your endpoint.
The Hermes v0.22.1 and v0.22.2 notes added Super 3.5 default parameters and put Nemotron on the NVIDIA NIM default model list.
What It Costs To Run
The model is free to download, and Hermes Agent is open source.
| Route | What you pay for |
|---|---|
| Bolt in Hermes Local Models | You pay for the hardware and the electricity, with no per-message fee. |
| Super VL through NVIDIA's hosted NIM | You use an NVIDIA API key, and NVIDIA's own terms decide the cost. |
| Super VL through a local NIM | You pay for the DGX or GPU hardware you run it on. |
I'm not quoting prices here, because hardware and API prices change fast and depend on where you buy.
Limits Before You Commit
The hardware bar is high, because NVIDIA's recipes ask for 32GB-plus RTX cards or 64GB-plus Spark machines.
Hermes support is brand new, and the v0.22.3 notes say this part of the release stays undocumented until v0.23.0.
The Hermes Local Models docs still say Hermes never offers builds below 4-bit, while Bolt is NVIDIA's own 2-bit build, so the docs and the catalog don't match yet.
The benchmark numbers in this post are NVIDIA's own, not independent tests.
On NVIDIA's speed test, Qwen3.6 35B-A3B and Gemma 4 26B-A4B ran faster than Bolt, so its selling point is intelligence per GB rather than raw speed.
I haven't run Bolt myself, so everything here comes from NVIDIA's model cards and blog, the Hermes docs, the Hermes catalog file and the official release notes.
I have run Hermes on a free local model, which I showed in my Gemma 4 video below.
🔥 Want help turning one local agent into a system that runs your business? The AI Profit Boardroom has the Agent OS, the Hermes training and four weekly coaching calls where you can bring your exact use case. → Join the AI Profit Boardroom
Who Should Use Nemotron 3.5 Bolt In Hermes
Businesses with a strong NVIDIA workstation and private client data should try it first.
Agencies that want an always-on local agent for research and admin should look at it closely.
Small teams on ordinary laptops should start with a smaller local model or use Super VL through NIM.
If you're still choosing a setup, my guide to the best Hermes Agent setup compares where your agent can live, and my best Hermes Agent model guide compares the brains.
Related Reading
- My best Hermes Agent setup guide helps you decide where your agent should run.
- My best Hermes Agent model guide compares the main brains for Hermes.
- My Hermes Agent Microsoft Store guide shows how a business can put Hermes on Windows 11 PCs.
Also On Our Network
- 🌐 The four-step Nemotron 3.5 Bolt Hermes setup walkthrough
- 🌐 Every Nemotron 3.5 Bolt spec, checked and scored
- 🌐 How Nemotron 3.5 Bolt fits an Agent OS workflow
- 🌐 Whether any Nemotron model is on GoldieBench yet
FAQ: Nemotron 3.5 Bolt Hermes For Business
What is Nemotron 3.5 Bolt in Hermes?
It is NVIDIA's Nemotron 3.5 Bolt 74B-A9B 24GB model running as the local brain of Hermes Agent.
Hermes added it to its local runtime in v0.22.3 on 21 October 2026.
Can a business use it commercially?
NVIDIA's card says it's ready for commercial or non-commercial use, governed by the OpenMDW License Agreement, version 1.1.
Read that licence yourself before you build a product on it.
What does a business need to run it?
NVIDIA's GGUF card lists recipes for RTX Spark machines with 64GB or 128GB and RTX cards with 32GB or 48GB and up.
It lists NVIDIA Ada Lovelace or Blackwell GPUs on Windows or Linux.
Does it send my data to the cloud?
Not when it runs in Hermes Local Models, because the docs say local models run entirely on your own machine after the download.
What if my hardware can't run Bolt?
The Hermes docs show how to use Nemotron 3.5 Super VL through the NVIDIA NIM provider instead.
If you'd like my team to show you how AI agents and content can win your business more Google and AI search traffic, book your free SEO strategy session here.
📺 Video notes + links to the tools 👉
🎥 Learn how I make these videos 👉
🆓 Get a FREE AI Course + Community + 1,000 AI Agents 👉
About Julian
I'm Julian Goldie, an SEO entrepreneur, author and founder of the AI Profit Boardroom, which has 3,400+ members.
I help business owners scale with AI agents, automation and SEO.
- I've built a 7-figure SEO and link-building agency, Goldie Agency, with 70+ people in Bangkok.
- I've grown a YouTube channel to 400,000+ subscribers.
- I wrote the book "Link Building Mastery", which covers nine link building strategies.
- I run GoldieBench, where I score AI models on real one-shot builds.
→ Get my best AI training inside the AI Profit Boardroom
Start with one private task on one strong NVIDIA machine, and you'll see quickly whether the Nemotron 3.5 Bolt Hermes setup belongs in your business.
