⭐ Get the Agent OS + join 3,400+ founders inside the AI Profit Boardroom → Join AIPB ($49/mo)

GoldieBench Blog · 9 min read

Hermes Agent Plugins Marketplace: What It Is, How To Use It, And Which Brain To Run It On

Hermes Agent plugins marketplace explained, plus which GoldieBench-scored models to run Hermes on once plugins add tools: Opus 5 8.27, Kimi K3 7.89.

Hermes Agent Plugins Marketplace: What It Is, How To Use It, And Which Brain To Run It On — illustrated hero

The Hermes Agent plugins marketplace is Nous Research's official Plugin Catalog, where you browse free plugins for Hermes Agent and install them into Hermes Desktop with one click.

Hermes Agent is the free, open-source AI agent from Nous Research, and Hermes Desktop is its app for Mac, Windows and Linux.

Every plugin in the catalog is free to install, and the catalog itself launched in mid-September 2026.

When I ran hermes plugins browse on 7 October it listed 469 curated plugins, with 9 official and 460 from the community.

It solves the old problem of extending Hermes by hand, where I once spent about three hours on GitHub for a skill that didn't even work.

In this post I'll explain what the marketplace is, how it works, how to install a plugin, every plugin I showed in my video and the safety checks to run first.

What the Hermes Agent plugins marketplace is

The Plugin Catalog lives on the official docs site at hermes-agent.nousresearch.com/docs/plugins.

A plugin is an add-on that gives Hermes something new, such as a desktop page, a search tool, a chat platform or a voice.

Every listing shows stars, a category, the maintainer, when it was added and last updated, a version where the author has tagged one, the GitHub repo and the README.

Every listing also has an "Open in Hermes Desktop" button and the matching terminal command.

In my video I called it 500 free plugins, and 469 curated entries on 7 October is close to that.

How the Hermes plugin catalog works

Every page in the Hermes Agent plugins marketplace follows the same layout, and reading it properly takes about thirty seconds.

The star count is your fastest quality signal, so compare it against other plugins in the same category.

The added and updated dates tell you whether the maintainer is still active, which matters because Hermes ships updates fast.

The version number tells you whether the author has tagged a proper release.

The "Reviewed source" link shows the exact commit the catalog pinned, so you know which code you are installing.

The README tells you what the plugin does and what it needs from you, such as an API key.

Hermes Tailscale is a good example, because it showed 52 stars, version 0.0.6 and an update on 4 October 2026 when I checked.

You can filter the whole catalog by source, which means official or community, and by category.

The categories cover desktop, tools, memory, models, automation, web and browser, platforms, voice and general.

When you install a plugin, Hermes runs a security scan first and refuses anything marked dangerous.

How to install a plugin from the Hermes plugin catalog

This is the one-click flow I showed in the video.

  1. Open the plugin's page and read its README first.
  2. Click "Open in Hermes Desktop", and Hermes Desktop opens an install dialog.
  3. Choose where the plugin appears, such as in the desktop interface.
  4. Check the GitHub repository the dialog shows you.
  5. Choose which Hermes profile gets the plugin, then click Install.

The terminal route still exists, and I confirmed these command names with hermes plugins --help on my own install.

CommandWhat it does
hermes plugins searchIt searches the curated catalog.
hermes plugins browseIt lists every curated plugin with its category and tier.
hermes plugins infoIt shows the repo, pinned commit and what a plugin adds, including any tools or hooks.
hermes plugins installIt installs a plugin after a security scan.
hermes plugins enable / disableThey switch a plugin on or off without removing it.
hermes plugins capabilitiesIt shows what each installed plugin declared and what it was granted.

Old way versus new way

The old way of extending Hermes was the terminal, cloning repos, editing config files by hand and restarting.

A broken install was hard to spot, and it could cost you hours before Hermes ever saw a new tool.

The new way installs in one click from a page that already shows you the stars, the repo and the pinned commit.

That is the real upgrade, because the catalog makes adding a plugin cheap and leaves your time free for testing whether it actually helps.

Every plugin from the video, and what it does

Here is every plugin I showed in the video, with the star counts I saw on 7 October.

The last column shows whether a plugin hands your model a new tool to call, which matters for the model section further down.

PluginWhat it doesStars on 7 OctDoes it lean on the model?
Hermes NewswireA scrolling news ticker at the bottom of a new chat, with sources, delete, resize, on/off, fast/slow and pause settings.37No, because the catalog says it uses zero model tokens.
Hermes TailscaleIt connects you to Hermes across devices, such as your phone and laptop.52Not much, because it is a desktop page.
WhatsApp agent platformIt connects Hermes to Meta's WhatsApp Agent Platform API.10Yes, because every chat becomes a model turn.
Minecraft gatewayIt puts a Hermes bot into a Minecraft server and relays chat.1Yes, because every player message becomes a model turn.
Brave SearchIt adds a brave_web_search tool.3Yes, because the model decides when to search.
Ollama Cloud web searchIt adds web search and page extraction.1Yes, because the model decides when to search and extract.
Voice plugins such as Hermes SpeechThey give Hermes a voice so it feels like Jarvis.196 for Hermes SpeechPartly, because speech feeds the model's turns.
Job searchIt filters job postings on plain-English preferences and writes digests.50Yes, because it judges postings against your preferences.
Hermes OfficeIt gives your Bot Mode agents a visual office, and I called it Hermes Office 2 in the video.16Indirectly, because the bots doing the tasks run on your model.

Desktop add-ons like Newswire, Tailscale and Hermes Office change what you see, while search, chat and job plugins change what your agent can do.

The safety check that matters most

This is the one thing I told everyone to check in the video.

Most plugins are community-made, so read the GitHub repository and README before you install anything.

hermes plugins install runs a security scan and refuses anything marked dangerous.

The Newswire README explains that a "caution" result becomes a confirm prompt in the terminal, while the Desktop install refuses it and lists the findings.

Even the strongest model will happily call a badly written tool, so a good brain is no excuse to skip the README.

Limits: don't install everything

Every tool you add is one more option your agent weighs on every turn.

A smaller brain with forty plugins can get clumsy, while the same brain with three plugins has far fewer ways to go wrong.

Pick two or three plugins that solve a real problem, test them in a spare profile, and add more only when you need them.

Does the Hermes Agent plugins marketplace come with its own model?

No, it doesn't, because the marketplace only adds plugins and Hermes Agent runs on whichever model you plug into it.

Hermes Agent is the body, and the model you choose is the brain that decides when to use every new tool a plugin adds.

I keep one Hermes profile per model, so I can switch brains and compare them on the same plugins.

To choose a brain, I use GoldieBench, the leaderboard where I score models on real one-shot builds.

I want to be straight about what the board measures.

GoldieBench gives every model the same one-shot prompt to build a game, a page, a simulation or a visual, then scores the rendered result from 0 to 10.

It does not run a dedicated tool-calling test, so these numbers tell you how capable a brain is at hard multi-step work, not how it scores on a plugin benchmark.

I still use the board as a guide, because a model that holds a hard multi-step build together is usually the safer bet for a crowded tool list, but you should test it on your own plugins.

Which brain to run Hermes on once plugins add tools

ModelLive GoldieBench avgDemos scoredHow I'd use it under Hermes
Claude Opus 58.2750The strongest single brain on the board for long, tool-heavy plugin stacks.
Hermes MoA8.1747A Hermes Mixture of Agents panel, which shows how far a Hermes system can go when several brains work together.
GPT-5.6 Sol8.1650A frontier alternative to Opus at the same $5 per million input tokens.
Qwen 3.88.1045A near-frontier score from a model that runs on a Qoder plan.
MiniMax M37.9747The value pick at $0.30 per million input tokens with a 1M context.
Kimi K37.8950My kimi-k3 Hermes profile for long unattended agent runs.
GLM-5.27.7747An open-weights brain with a 1M context for long memory replays.
Claude Opus 5.57.5750The model behind my Claude-subscription Hermes profile.
Grok 4.77.1520A solid Grok option, scored on 20 played games.
Qwable 5 27B Coder7.1441The best free local brain on the local board, but it runs at about 20 tokens per second.
Laguna XS 2.13.9342A free model for light jobs, not for a big plugin stack.

Every number above matches the live model page when I checked it on 7 October 2026.

The pattern is simple, because the more tool and platform plugins you add, the more decisions your brain makes on every single turn.

My routing rule for a plugin-heavy Hermes

If your plugins are mostly desktop add-ons like Newswire, Tailscale and Hermes Office, your brain choice barely matters, because those plugins hardly touch the model.

If you add search tools, a WhatsApp channel or a job-search agent, move up to a brain from the top of the board.

For long unattended runs with lots of tools, I'd use Claude Opus 5 or Kimi K3.

For bulk work on a budget, I'd use MiniMax M3 or GLM-5.2.

For private, offline work, I'd use Qwable 5 27B and keep the plugin list short, because a slower local model feels every extra tool.

The free cloud models are fine for testing a plugin, but Laguna XS 2.1 at 3.93 shows why I wouldn't hand one a big tool stack.

If you want the full brain ranking, my post on the best LLMs for Hermes Agent goes deeper, and the best local model for Hermes Agent covers the offline lane.

How I test a new plugin against my Hermes brain

Here is the routine I use before a plugin goes anywhere near my main profile.

First, I install it into a spare profile, so my main setup stays untouched.

Second, I run hermes plugins info to see whether it adds tools, hooks or environment variables.

Third, I give the agent three normal jobs that should use the new tool and three jobs that shouldn't.

Fourth, I watch whether the brain calls the tool at the right moments and leaves it alone the rest of the time.

Fifth, I swap the profile to a cheaper model and run the same six jobs again.

If the cheaper model handles the plugin cleanly, I keep it there and save money.

If it starts calling tools at random, I move that plugin's work back up to a stronger brain from the table above.

That small test tells you more about your own plugin stack than any single number on a leaderboard.

Pick a brain that has earned its place on the board, add plugins one at a time, and you'll get the most out of the Hermes Agent plugins marketplace.

FAQ

What is the Hermes Agent plugins marketplace?

It is the official Hermes Agent Plugin Catalog from Nous Research at hermes-agent.nousresearch.com/docs/plugins. It lists free official and community plugins with stars, categories, dates and the GitHub repo, and most install with an Open in Hermes Desktop button.

How many plugins are in the Hermes plugin catalog?

When Julian ran hermes plugins browse on 7 October 2026, the curated catalog listed 469 plugins, made up of 9 official and 460 community plugins.

Does the Hermes Agent plugins marketplace come with its own model?

No. The marketplace adds plugins, and Hermes Agent runs on whichever model you connect to it. Julian keeps one Hermes profile per model so he can switch brains and compare them on the same plugins.

Which model should I run Hermes Agent on when I use plugins?

For tool-heavy plugin stacks, Julian picks a top-board brain such as Claude Opus 5 (8.27 on GoldieBench) or Kimi K3 (7.89). For budget work he uses MiniMax M3 (7.97) or GLM-5.2 (7.77), and for local work Qwable 5 27B (7.14) with a short plugin list.

Does GoldieBench test tool calling for Hermes plugins?

No. GoldieBench scores one-shot builds such as games, pages and simulations from 0 to 10. The scores show general capability on hard multi-step work, which Julian uses as a guide when choosing a Hermes brain.

Are Hermes plugins safe to install?

Most are community-made, so read the GitHub repository and README first. The installer also runs a security scan, refuses dangerous plugins and asks for confirmation on caution results in the terminal.

The same stack Julian uses

Run this stack yourself.

Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 3,400+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,400+founders
258documented wins
38countries
$49/momonthly