Hermes Agent’s v0.22.0 release adds guided setup, connectors for popular apps and one-click local model support, alongside a much larger set of changes to the agent. For people who want to run an AI assistant themselves, the standout is a simpler path to trying it without an API key or sending model work off their own hardware.
Nous/Hermes Watch analysis
What happened
Nous Research’s v0.22.0 release notes describe a release dated 13 October. A fresh desktop install now opens with a guided setup chat that runs on a free trial model. The release also adds connectors for apps including Gmail, Linear and Notion, and a local-model setup that installs and manages llama.cpp, checks what fits the user’s GPU and RAM, and runs the model on their machine.
There is plenty beyond the front door. The release includes a catalogue of 600 reviewed plugins, browser sign-in using saved credentials from 1Password, Bitwarden or Hermes’s encrypted vault, and a Bot Screen that lets someone take over when an agent hits a login, two-factor prompt or CAPTCHA. Nous says the model does not see the saved password, and card or address fills at checkout require confirmation.
Our top picks
- Local models without the usual setup tangle
Hermes selects a model that fits available GPU and RAM, and supports CPU, CUDA, Metal, Vulkan and ROCm. - Connectors for everyday apps
Users can connect apps such as Gmail, Linear and Notion, with no Nous account needed to start. - A gentler first launch
New desktop users get a guided setup chat and can try Hermes before they have an API key. - A 600-plugin catalogue
Plugins are pinned to reviewed commits, with catalogue rules restricting changes to Hermes internals and unscannable binaries. - Human takeover for browser roadblocks
Bot Screen lets users handle logins, two-factor checks and CAPTCHAs, then hand control back to the agent. - A browser password-manager link
Hermes can use saved logins without showing the password to the model; checkout details still need user confirmation.
Why it matters
This release tackles the awkward parts of using an agent: getting started, connecting it to useful services, choosing a model that will run locally and stepping in when a website wants proof that you are a person. The result is a more practical route from installing Hermes to putting it to work, with local inference available for people who prefer to keep model activity on their own hardware.
Some changes deserve a glance before upgrading. Managed installs now use Python 3.14; the Windows desktop is MSIX-only and requires Windows 11 22H2 or newer; default macOS builds are for Apple Silicon. Hermes also enables automatic pruning of ended sessions inactive for 90 days, though pinned and open sessions are excluded and the setting can be disabled.
Our read
This is a substantial step towards making Hermes easier to try and more useful across desktop, browser and local-model workflows. The best bit is not any single grand promise: it is that setup, integrations and running a model locally are becoming parts of one product rather than separate weekend projects. Check the platform and session-pruning changes before upgrading; convenience is less charming when it quietly deletes something you meant to keep.
What to watch
- Whether the new connectors and local-model setup work reliably across supported platforms and hardware.
- How the reviewed plugin catalogue grows, and how its safeguards hold up in use.
- Whether users find the new default session-pruning behaviour clear enough to manage.
Discussion spark: Would you trust an AI agent with app access and saved logins if you can take over at sensitive steps, or should those connections stay manual?
Sources and evidence
- Hermes Agent v0.22.0 – The Threshold Release (14 October 2026, 03:21 UTC)
Nous/Hermes Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Nous Research.