Skip to content
HomeNewsHermes Agent 0.23 brings every chat under one roof

Subscribe to our newsletter

One short email when we publish. No spam, unsubscribe anytime.

Hermes Agent 0.23 brings every chat under one roof

Hermes Agent 0.23 brings every chat under one roof

By OneClickClaw EditorialNovember 7, 2026

Nous Research released Hermes Agent 0.23, which it calls The Hearth Release, on the 5th of November 2026, according to the release notes on GitHub. The notes lead with a browser version of the desktop app. Underneath it, every conversation now lives in one place: a chat started in the terminal, the desktop app, a messaging bot or a scheduled job keeps one queue and one history.

OneClickClaw ran the release through its own checks on the 6th of November. A new server on version 0.23 was ready in about seven minutes and answered its first test message within seconds. An existing server moved up from the previous version in about three and a half minutes and answered a test question correctly. From that day, every new Hermes Agent server on OneClickClaw starts on 0.23, and customers who already run Hermes pick the moment to move from their own dashboard.

The notes count about 10,080 changes to the code since version 0.22, from 666 contributors.

One home for every conversation

Hermes can be reached from many places: a terminal window (the text screen where commands are typed), the desktop app, a code editor, Telegram, or a job that runs on a schedule.

In 0.23, each machine runs one gateway, the part of Hermes that receives every message and hands it to the agent, and every window connects to it, the notes say. Each message is delivered once. A chat opened anywhere is the same chat, and a waiting message can be edited from the desktop app without running twice.

Rewind a conversation

Comic illustration of Hermes at a marble desk lit by an oil lamp, rolling a long scroll back to an earlier passage with a raised eyebrow.
Pick an earlier message, change it and send it again. The original conversation stays as it was.

In the terminal, pressing the Escape key twice opens a list of earlier messages, according to the notes. The owner picks one, edits it and sends it again, or starts a new conversation from that point. The original conversation stays exactly as it was.

Group chats that pick who answers

Bot Mode lets several Hermes bots share one room. In 0.23, a lead bot, or Hermes itself automatically, decides who replies to a message that names nobody. Scheduled jobs can post into a room, and every room keeps a purpose and shared notes.

A back and forth between bots stops by default after four turns following the last message from a person, the notes say. Each bot also shows what it is doing while it works, for example which tool it is using and for how many seconds.

A price on every paid tool

Comic illustration of Hermes on a marble terrace at night, weighing bronze coins on small balance scales with a sly half smile.
Paid tools now show their price before they run, with a running total for the chat.

Some add-on tools cost money each time they run, such as tools that create 3D models or media. In 0.23 these tools can state the price of each call. Hermes shows that price when it asks for approval, keeps a running total for the chat, and asks again before the chat goes past its spending limit. It asks even when the owner has told it to approve everything on its own.

Plugins that keep themselves current

A plugin is an add-on that gives Hermes a new skill or a new connection. One command now updates all of them, and an optional switch keeps them up to date on their own. Updates that would break a setup are held back, the notes say.

The plugin catalog that Nous Research runs grew from about 600 entries to more than 1,050. A new Verified label marks plugins built by the company behind the service they connect to. The built-in skills went from 58 to 44, partly because Word, PowerPoint and Excel now share one Microsoft Office skill. An owner can also lock a skill written by hand so the agent never rewrites it.

Built for an agent that never switches off

Comic illustration of Hermes standing awake on the Olympus summit at night, beside a marble water clock and a small eternal flame, looking toward the first light of dawn.
Scheduled jobs run on time on a machine that stays awake.

The notes describe 0.23 as lighter and faster, especially when Hermes runs all day:

  • Repeated runs of the same scheduled job share one prompt cache, a stored copy of the instructions that AI providers bill at a lower rate when it is read again.
  • A one-off task sends about 1,570 fewer tokens per call. A token is a small piece of text, and providers charge by the token.
  • A running scheduled job can be stopped, and each job can be told what to do with runs that fell due while Hermes was off.
  • A gateway that runs for weeks now frees the computer's working memory once a chat is closed, and the chat itself stays saved. The gateway stays responsive over weeks of running.
  • Old chats stay on the machine. Clean-up of past sessions is a setting the owner switches on, and Hermes asks once when the disk runs low.

On a laptop that sleeps, a job set for seven in the morning runs late or is skipped, depending on that setting. On a machine that stays awake, it runs at seven.

OneClickClaw runs Hermes Agent on its own always-on server, one per customer, kept running around the clock and now starting on version 0.23. The 7-day free trial needs no credit card, and the Starter plan for a Hermes Agent server costs €14.99 per month after that.

Source: github.com

Helpful documentation

Want your own Hermes agent running without the setup? We host it on a dedicated EU server. Free for 7 days, no credit card.

Start your free trial

How managed Hermes Agent hosting works

Get notified when we publish new articles

No spam, unsubscribe anytime.

By subscribing, you agree to our Privacy Policy.