Three announcements landed at once: Hermes Agent now installs in a single click, Nous Portal made Tencent's Hy3 AND poolside's Laguna S 2.1 free, and your chat database just shrank by up to 78%. This is the read-along: the tweets, what each one actually means, the setup, and how it all snaps into the Agent OS.

Read that spec line again: 118 billion total parameters, 8 billion active.
That's the MoE trick — frontier-scale knowledge with small-model speed.
We benched its little sibling, Laguna XS, on GoldieBench when it dropped. The S is the one poolside calls their most capable release ever — and for two weeks it costs nothing on Nous Portal.
What it's for: coding-heavy agent work. poolside builds coding models; this is the brain you point Hermes at for build tasks while it's free.
This one we know intimately.
Hy3 averaged 7.41 on GoldieBench — one of the strongest agentic coders we've ever tested, with skill-infused prompts taking its Doom build from 3.5 to 8.2.
Nous says it themselves: cost-effective agentic use, strong on coding, tool-calling reliability, reasoning and long context.
Translation: this is a Hermes-shaped model. Tool calls that don't fumble, long agent loops that don't lose the plot — free for another week.
Full build gallery: our Hy3 guide · scores on GoldieBench.
Unsexy. Massive.
Hermes stores every chat session in a local database, and heavy users watch it balloon.
The new optimization cuts it by ~60% on average, up to 78% — and if your database is under 1GB, the migration runs itself. Bigger ones get a prompt.
Why you care: faster session loads, snappier history search, and a lighter disk footprint on the machine your agent lives on 24/7.
Grab the Hermes desktop app, click install, done.
No terminal, no dependencies, no config files. The days of pip-install archaeology are over — this is the "my mum could install it" era of local agents.
Portal is Nous's hosted model service, and it's where both free models live right now.
Sign in from inside Hermes, open the model picker, and select Laguna S 2.1 or Hy3. That's the whole wiring.
My standing setup: a Portal free model as the main brain while the promo runs, a local model (Gemma 4 or Agents-A1 via Ollama) as the offline fallback and sub-agent.
When a free window closes, you switch models in the picker — the system never changes. That's the whole philosophy: system over models.
Agent loops + coding: Hy3 — its 7.41 came from exactly this kind of work.
Hardest reasoning while the window's open: Laguna S 2.1 — biggest brain on the free tap.
Offline, private, forever: your local lane — Agents-A1 or Gemma 4 via Ollama.
All three plug into the same Hermes. Switching is a dropdown, not a migration.
Before
Getting an agent running meant a terminal, a Python environment, and an evening you didn't get back.
Then paying API rates to give it a brain good enough to matter.
Most people quit at step one. I watched it happen in our community every week.
Then the stack collapsed into one click and a free model picker.
After
Now: install Hermes with a click, sign into Portal, pick a free frontier brain.
Five minutes from zero to a working agent — with Hy3, a model we scored 7.41, costing nothing.
The excuse era is over. The only question left is what you'll automate first.
And the Agent OS below is where it stops being a toy and starts running your business.
I've built my whole operation around this agent — the voice copilot, the SEO pipeline, the overnight kanban crews — and I bench every brain that plugs into it on GoldieBench, in public, with playable demos.
No invented quotes — the member wins are written by members, raw, in a 158-page doc.
Read the 158-page wins doc →The install is one click. The brains are free. The video above walks every step.
So here's the deal: before you sleep tonight, have Hermes installed with a free Portal brain selected, and give it ONE real job — a summary, a file cleanup, a draft.
The people who set it up this week, while the free windows are open, are the ones who'll actually feel what an always-on agent changes.
One click. One brain. One job. Tonight.
Everything in this guide — the one-click Hermes, the free Portal brains, the local fallback — ships pre-wired inside my Agent OS:
Say the wake word and Hermes answers, acts, remembers to your vault, and reads your morning briefing.
Keyword → five unique articles → deployed → indexed. The pipeline behind these rankings.
Drop a task; planner → builder → reviewer agents move it overnight.
Every agent reads and writes one vault — Hermes knows what Claude did, and your business context never resets.
One click installs Hermes. The Agent OS is what makes it a business system — and it's updated the same week every drop like this lands.
It's the operating system I run a seven-figure business on. You get the whole thing.
Get the Agent OS →This guide is literally the answer: right now the main brain is free (Hy3 or Laguna S on Portal), the fallback is a free local model, and the frontier work rides the CLIs you already pay for. Plus the Boardroom's token-minimisation playbook cuts usage further. A full day of agent work at $0 isn't a promise — it's the current setup.
Wrong: "Setting up an AI agent is technical."
Right: It's one click now. The hard version died this release. If you can install Spotify, you can install Hermes.
Wrong: "Free models are the weak ones."
Right: Hy3 scored 7.41 on our bench — frontier-class agentic coding — and it's on the free tap right now. The free tier has never been this strong.
Wrong: "I'll set it up when things settle down."
Right: The free windows are the settling. Hy3: one more week. Laguna S: two. The system you build during them runs forever — the brains just rotate.
158 pages of members running this exact stack — their words, their numbers.
Read the 158-page wins doc →One click for the agent. Two free frontier brains. A database that just got lighter. And a community that wires every drop like this into a working system the same week it lands.
Install Hermes tonight, then come get the full machine — the Agent OS, the daily tutorials, the coaching calls, and 4,000+ founders who'll answer when you're stuck.
Join the AI Profit Boardroom →Install it, pick a free brain, give it one job tonight — then check what the bench says at goldiebench.com. I'll see you in the next one.