Hermes Agent pet sprites turn agent status into a glanceable mascot

Your terminal agent can finally look busy without lying to you.

In mid-June 2026, Nous Research shipped pets for Hermes Agent: animated mascot sprites from the open petdex gallery that mirror what the agent is doing. Idle, thinking, running a tool, waiting on you, finishing clean, failing hard. Nearly 3,000 options. Off by default until you opt in.

The AlphaSignal digest called it cute. I call it peripheral status design for long agent sessions.

Why a sprite beats another spinner

Agent UIs already dump logs. Logs demand focus. A mascot rewards glanceability: you notice the jump animation when a plan completes while you are in another tab.

Hermes maps six states to sprite rows:

Nous merged the feature in PR #46464 across CLI, TUI, and desktop. Desktop adds a floating pet you can shift-click into an always-on-top window with status bubbles.

Cosmetic by design (and that is the point)

Pets do not touch prompts, skills, memory, or MCP servers. Documentation is explicit: no effect on tokens or caching. The renderer reads agent state events and picks a sprite row.

That matters for teams auditing agent spend. You can add delight without another hidden context injection.

Feature is dormant until you install and select a pet. Privacy-conscious teams can ignore it entirely.

Gallery scale and custom hatch flows

Petdex hosts 3,200+ community sprites (otters, dragons, robots, cats). Install flow:

hermes pets list cat          # filter gallery
hermes pets install boba --select
hermes pets scale 0.5         # clamp 0.1–3.0
hermes pets off               # disable without uninstall

Hermes stores pets per profile under <HERMES_HOME>/pets/, not the petdex npm path under ~/.codex/pets. Use hermes pets, not the standalone petdex CLI, to avoid path confusion.

You can also hatch custom pets:

• /hatch or hermes pets generate flow in CLI
• Desktop gallery with size slider in Settings → Appearance
• Reference-image grounding through Nous Portal, OpenRouter, or Krea so each animation row keeps the same character

Generated sprites pack into the standard 8×9 petdex atlas (192×208 cells). Valid output can be submitted back to petdex.

Terminal rendering details builders notice

Full fidelity needs a graphics-capable terminal: kitty, Ghostty, WezTerm, iTerm2, or sixel. Otherwise Hermes falls back to truecolor Unicode half-blocks. Pipes and redirects disable terminal rendering (no TTY), which is the right default for CI logs.

hermes pets doctor diagnoses render mode mismatches before you blame the sprite pack.

How this fits the Hermes stack I already track

Hermes Agent crossed 180k GitHub stars in 2026 with serious infra: persistent memory, skills loops, messaging gateways, Herald voice. Pets sit in the Media & Web section next to Spotify and TTS, which is honest product placement.

If you read my posts on Hermes v0.20 Herald A2A voice or scheduled agent prompts, pets are the opposite axis: zero backend complexity, pure operator UX.

For client work I would not pitch pets as ROI. I would pitch them as on-call fatigue reduction when engineers run three parallel agent sessions and need to know which one finished without reading stderr.

When I would skip it

Air-gapped installs with strict terminal policies may block graphics protocols.

Screen reader primary workflows get little from sprite animation (status text still exists in logs).

Brand-sensitive ops consoles may prefer plain status chips.

Takeaway

Hermes pets are the rare feature that admits it does nothing to model quality and still ships. In a market obsessed with autonomy scores, Nous bet on human situational awareness. Pick a sprite, scale it down, pop it out on desktop, and stop alt-tabbing to check whether the agent is stuck or done.

Building agent UX for operators? Get in touch if you want help wiring status surfaces that actually reduce pager noise.