Here’s the honest rundown. Hermes agent Quicksilver — the V0.19 release — is all about speed: replies land about 80% faster, agents finish jobs even if the app crashes, and one gateway now runs a whole fleet of agents.
I updated the day it shipped. Here’s exactly what changed and how to get the most out of it.
Last updated: July 2026.
Key takeaways
- Quicksilver (V0.19): first word in under a second — cold start cut from 4.3s to 0.9s (~80% faster).
- Jobs and replies are written to a database, so agents finish even if the app or gateway crashes.
- One gateway routes to a fleet of agent profiles — each with its own model, memory, skills and secrets.
- Smart approvals: a second AI assesses flagged commands, with deny rules that hold even in full auto.
- Two new free models supported: HY3 and Laguna S 2.1. Get the ready-made setup in the AI Profit Boardroom.
What Is the Quicksilver Release?
Hermes is the free, open-source agent from Nous Research — it lives on your computer, talks to any model you point it at, and does real work. V0.19 is codenamed Quicksilver, and the name is literal: the team rebuilt how fast it wakes, how fast it answers, and how safely it runs many agents at once. It shipped 2,245 commits in a single release.
Before this, agents felt like hiring the world’s fastest assistant and making them sit in a waiting room: 4–6 seconds of blank screen, background jobs silently dying if the app crashed, and every risky command needing a manual click. Quicksilver fixes all three.
The Three Core Upgrades
| Upgrade | Before (V0.18) | Now (V0.19) |
|---|---|---|
| Speed | 4.3s cold start, blank screen | 0.9s cold start, first word <1s, live thinking notes |
| Reliability | Crashes killed background jobs | Jobs + replies written to a database — they finish and deliver |
| Fleet routing | One agent, one brain for everything | One gateway routes channels to separate specialist profiles |
Fleet Profiles: One Gateway, Many Specialists
This is the big structural change. Each profile is a fully separate agent — its own model, memory, skills and secrets — and the gateway routes messages to the right one. So you can run an SEO profile, a writing profile and an ops profile, and your main Hermes auto-delegates between them.
It also solves rate limits: put HY3 (new free model) on one profile, Laguna S 2.1 (also free) on another, a local model on a third — and tasks route around whichever is slow or limited, automatically. No babysitting.
Smart Approvals: Speed With Brakes
If you’ve ever been nervous leaving an agent running, this is for you. The default is now a third way between manual-approve-everything and full auto: when a command gets flagged, a second AI independently assesses it before it runs. You can still choose manual or full auto — and your own deny rules hold even in full auto mode.
You get the speed of autonomy with brakes you control.
The Smaller Upgrades Worth Knowing
- Real secrets management — keys out of plain-text configs
- Session export to Markdown/HTML, with a redact flag that scrubs secrets first
- /subscription — billing info without the dashboard
- Safe mode — boot Hermes with all customisations off if your config breaks
- Reasoning dials and stat skills
- Nicer buttons on Telegram, Discord and Matrix
How to Update
Easiest route: in my Agent OS, go to Manage and click Update Hermes — done. Or run hermes update in the terminal. Everything downstream gets faster too: the Kanban board delegates quicker, the memory galaxy retrieves context faster, and workflows like Apollo, Oracle and Astros all feel snappier. See my Hermes voice activation guide for the other new release, and agentic OS for Hermes for the wider setup.
Get the Ready-Made Setup
The Agent OS with Quicksilver, the fleet profiles, free models and all my workflows pre-configured is inside the AI Profit Boardroom.
New here? Start free with my AI Money Lab community (free AI course + 1,000+ AI agents), or grab a free strategy session.
FAQ
What is Hermes agent Quicksilver?
The V0.19 release of the free, open-source Hermes agent — focused on speed (80% faster replies, 0.9s cold start), crash-proof jobs, and one gateway running a fleet of specialist agent profiles.
How much faster is it?
Cold start dropped from 4.3 seconds to 0.9 — about 80% — and the first word now lands in under a second, with live thinking notes while it loads.
What are fleet profiles?
Fully separate agents (own model, memory, skills, secrets) behind one gateway, which routes each message to the right specialist — and auto-delegates between them.
Is it safe to run in auto mode?
Safer than before — a second AI assesses flagged commands by default, and your deny rules hold even in full auto.
What free models does it support?
Two new ones this release — HY3 and Laguna S 2.1 — plus local models. Spread them across profiles to dodge rate limits.
The Bottom Line
Faster wake, safer autonomy, specialist fleets — that’s Quicksilver. The setup is in the AI Profit Boardroom.