Nous Research said on September 23 that Hermes Desktop can stream an agent’s screen in real time, allowing users to watch a bot work during a session. The company announced the feature in a post on X.
Users can watch an agent operate a browser, type in a terminal or open a window, according to the post. They can take over keyboard and mouse control to interact with the screen or enter credentials, then return control to the bot.
The feature, called Bot Screen, assigns each Hermes agent a virtual desktop running Xfce on TigerVNC. That desktop streams to Hermes Desktop, giving users a view of an agent’s activity and a way to intervene in tasks such as logins or CAPTCHAs.
When the bot is controlling the session, the interface shows “Bot is in control,” the documentation says. Clicking Take over changes the border to red and gives the user keyboard and mouse control. Clicking Hand back restores the bot’s control. While a user has control, the bot’s computer_use and browser tools are refused with the message “human_has_control.”
The setup requires a Linux gateway host, rather than macOS or Windows, with TigerVNC, Xfce and Computer Use enabled. Nous estimates that each open screen with a browser uses roughly 1.1 GB to 1.5 GB of memory. The company also cautions that bots share the host user account, files and network, and that the takeover screen uses a US keymap.
Nous describes Hermes Desktop as an app for Mac, Windows and Linux within a product line that also includes Hermes Agent, Nous Portal and business and enterprise offerings. The open-source AI lab was established in 2023 by CEO Jeffrey Quesnelle, Karan Malhotra, Teknium and Shivani Mitra. In April 2025, Paradigm led a $50 million Series A in Nous, according to The Block.
