Home / Vault & Second Brain / The Morning My AI Got 80% Faster — Hermes v0.19.0 Quicksilver

The Morning My AI Got 80% Faster — Hermes v0.19.0 Quicksilver

The Morning My AI Got 80% Faster — Hermes v0.19.0 Quicksilver

The Morning My AI Got 80% Faster

Last Monday, I woke up to a notification that Hermes Agent v0.19.0 — the “Quicksilver” release — was out. I skimmed the release notes, saw “80% faster first-token time,” and thought: that cannot be real. Performance improvements are usually 10-15%, maybe 30% if you are lucky. 80% sounds like marketing. But I updated anyway, because I have learned that ignoring Hermes releases is how you end up with a broken setup three weeks later.

I was wrong. It was real.


What 80% Faster Actually Feels Like

Before the update, every time I asked Hermes something — a question about my vault, a command to check my Home Assistant setup, a request to write a blog post — there was a noticeable pause. About 4.5 seconds of silence while the model loaded, the context was assembled, and the first token trickled out. I had gotten used to it. I would type my question, look away, wait for the response to start appearing.

After the update, that pause is gone. The response starts appearing in under a second. It does not sound like much on paper, but in practice it changes the entire feel of the interaction. It went from “asking a question and waiting” to “having a conversation.” The difference between sending a letter and talking to someone in the same room.

I did not expect a software update to change how I perceive my own AI, but it did. The speed makes it feel less like a tool and more like a collaborator.


The Curator Guardrail That Saved My Skills

There was another change in v0.19.0 that I almost missed, and it is arguably more important than the speed improvement. The release introduced a “curator manual-skill guardrail” — PR #66579. What it does is simple: skills that were created manually (by me, by URL import, or from Gitee) are no longer automatically archived by the curator.

Why does this matter? Because I have several hand-crafted skills that I rely on every day: the vault-content-creator that publishes these blog posts, the genealogie-webapp that helps me research my family tree, the mem0-digest that summarizes my AI’s memory. Before this guardrail, the curator could decide these skills were “inactive” and archive them. I would wake up to a broken cron job and spend an hour figuring out why.

This fix is a good example of why I keep running the latest Hermes version. The developers at Nous Research are not just adding features — they are fixing the sharp edges that real users hit. Someone must have lost a manual skill to the curator, filed an issue, and the team fixed it. That is the kind of development cycle I can get behind.


What Else Changed

The release was bigger than just speed and guardrails. Over 2,200 commits, 1,000 merged pull requests, 450+ community contributors. Some highlights that affect my daily use:

  • Smart Approvals — Hermes now asks for confirmation before risky actions (file writes, commands, API calls). It is on by default. I was worried this would slow me down, but it actually caught me once already when a command was about to overwrite a config file I did not mean to touch.
  • Headed browser mode — The browser no longer closes after every turn. This sounds minor, but if you have ever tried to debug a web automation that kept disappearing, you know how frustrating that was.
  • Delivery ledger — Messages no longer get lost if the gateway crashes. This is one of those infrastructure changes you do not notice until it saves you, and then you wonder how you lived without it.
  • Live subagents — Real-time tracking of what subagents are doing. Useful when I run multi-agent workflows for research tasks.

Should You Update?

If you are running Hermes Agent, yes. The update is straightforward — pip install --upgrade hermes-agent — and the performance improvement alone is worth it. The curator guardrail is a bonus if you have manual skills. Smart Approvals might take a day to get used to, but you can disable them in config if they get in your way.

I have been running v0.19.0 for two days now. No regressions, no broken skills, no surprises. Just a faster, more stable AI that feels less like a tool and more like a partner.


Have you updated to Quicksilver yet? What was the biggest change you noticed? Drop a comment below — I am curious if the speed improvement hit you the same way it hit me.

Wat vond je van dit bericht?

Leave a Reply

Your email address will not be published. Required fields are marked *