Hermes Agent gets 3,200+ animated pet mascots

Nous Research added animated pet mascots to Hermes Agent, providing access to more than 3,200 sprite options drawn from the open-source petdex gallery. The mascots are purely cosmetic and do not change prompt processing, token usage, caching, or agent execution.

Each mascot maps to six agent activity states: idle when no action is running, running when a tool executes, thinking during model review, a wave when a turn ends cleanly, a jump when a full plan completes, and a failed pose when a task breaks.

The pets appear in the command-line interface, the terminal user interface and the official desktop application. Users can detach a mascot into a transparent, always-on-top window so the animation remains visible while the rest of the interface is minimized.

Installation and selection are command-line driven. The documentation shows commands such as hermes pets install boba –select to install and activate a sprite and hermes pets select <name> to switch mascots. The feature is off by default and controlled by the Hermes configuration setting display.pet.

Rendering depends on the environment. Terminals that support graphics protocols like kitty, Ghostty and WezTerm display sprites at full fidelity. On terminals without those protocols, Hermes falls back to truecolor Unicode half-block rendering so the mascot is still visible at lower resolution.

The desktop app renders the mascot on a canvas and shows short status messages such as “working…” and “thinking…” The pop-out window also displays an icon when tasks finish while the user is away.

Pet sprites come from the petdex gallery, which accepts community submissions under an open-source model. Nous Research published the feature in the Media & Web section of the Hermes documentation, and the gallery currently contains more than 3,200 sprites.

Hermes Agent launched in February 2026. The project has more than 180,000 stars on GitHub. Nous Research has developed other Hermes features including persistent memory, a self-improving skills loop, multi-agent support and messaging gateways for Telegram, Discord and Slack.

The documentation emphasizes that mascots provide visual feedback only and do not affect caching, tokens or the agent’s decisions.