Published: 2026-10-15
Summary

Hermes Agent v0.22, the Threshold release: one-click local models, a first-launch setup chat, connectors and password-manager logins

Chapters / key moments (click to jump — plays here on the page)

Hermes Agent v0.22, named "The Threshold Release", is about getting started and staying running: it can now run open models on your own machine without an account or API key, sets itself up in a chat on first launch, connects to hosted apps, and signs in to websites without showing the model your password. This summary follows a video from a channel that sells a community around Hermes, and checks each claim against the official release notes; where they differ, the notes win.

Source video

"Hermes Agent V0.22 Just Changed AI Agents Forever!" by Julian Goldie SEO — Watch on YouTube → The video includes repeated community pitches, which are not part of the release.

What the release notes confirm

These points match the v0.22.0 release notes (dated October 13; the GitHub release was posted October 14).

Local models. Hermes "installs and maintains llama.cpp with the right engine for your hardware" (CPU, CUDA, Metal, Vulkan or ROCm). The catalog includes Qwen3.8 27B and DeepSeek V4 Flash, plus any GGUF file from Hugging Face. In a terminal, run hermes model and choose Local models. The video adds that it scores each model against your memory and graphics chip, picks one that fits, lets downloads pause and resume, and updates the engine in one click.
First launch. A fresh desktop install opens a guided setup chat on the free nous/welcome trial model, which needs no API key. It is available once per operating-system user. In the video it asks your name, a colour, layout, which apps and plugins to connect, then offers a short tour and up to four suggested tasks.
Connectors. The agent can act in hosted apps such as Linear and Notion through the Nous tool gateway, and you don't need a Nous account to start. Manage them with hermes connectors or /connectors; upgrade the free identity to a full account with hermes auth upgrade or /login. The video says the desktop connectors page replaces the old MCP tab and lists hosted apps and your MCP servers together, and that you can keep a work and a personal account for the same app side by side.
Password-manager logins. The browser agent can sign in using 1Password, Bitwarden or Hermes's own encrypted vault, and the model never sees the password. Card and address fills at checkout need your confirmation. The video adds that one-time codes are filled without entering the conversation.
Plugin catalog. 600 reviewed plugins, each pinned to a reviewed commit; the video says the default install got lighter because extras moved into the catalog.
Long jobs. Prior reasoning is replayed by default on every provider that supports it, including across resume, branches and compaction. Transient 503, 529 and timeout errors are waited out in bounded cycles (agent.auto_recovery_cycles, default 5) instead of ending the turn.
Bot Screen. A bot on a Linux gateway can have its own desktop, which Hermes Desktop streams; you can take over for logins, 2FA or CAPTCHAs and hand back. It is off by default.
Stable channel. Hermes now publishes tested SemVer releases and source installs follow them by default. To track every change on main, run hermes update --set-channel main. The desktop app is a signed package that bundles its own Python and Node, with a signed MSIX for Windows (Microsoft Store) and a signed, notarised Apple Silicon build for macOS.

Other changes worth a look before you update

The release notes also list commands and defaults that change behaviour, none of which the video covers: /new or /reset start a fresh chat (idle and daily resets are retired), /model X --reasoning high sets effort when switching models, hermes gateway migrate --multiplex moves you to the single multiplexed gateway, hermes mcp login <server> --flow device does device-code login for MCP servers, and sessions are auto-pruned after 90 days by default (sessions.auto_prune). Read the notes before updating a production gateway.

Updating

The video's route: update from your dashboard, or open Hermes Desktop, check for updates and click update. It also mentions a small v0.22.1 follow-up that fixes the Windows Store build and puts a short list of recommended models at the top of the model picker; we didn't check that against the notes.

Gotchas & Caveats

The release notes don't state the free trial's usage limits, so don't build a routine on nous/welcome.
Bot Screen is off by default and the notes describe it for Linux gateways only.
Contributor and change counts in the video (about 2,500 people, 24,200 changes, 7,180 problems closed) are not in the portion of the notes we read.
A local model still needs enough memory; on small machines expect slower, weaker answers than a hosted model.

Key Takeaways

You can try Hermes with no account or key, and run open models through its built-in llama.cpp engine.
Connectors and password-manager logins make browser jobs possible without pasting secrets into chat.
Pin your update channel deliberately: stable by default, main only if you want every change.