explainx / blog
Deep tutorial for OpenAI Codex desktop pets: Settings, /pet overlay, hatch-pet install, sprite prompts, troubleshooting, security. Plus archetypes for best Codex pets and links to explainx.ai prompt kit & hub.
May 2, 2026
What shipped in Codex’s agent UI, how custom pets are packaged through OpenAI’s hatch-pet skill, and why a little dock-side animation can still be a serious product bet.
May 30, 2026
On May 29, 2026, OpenAI launched computer use for Codex on Windows, letting AI control Visual Studio, Excel, and real workflows while you steer from your phone. With self-managing threads, parallel worktrees, and usage stats tracking, OpenAI is closing the Windows gap and escalating competition with Anthropic's Claude Cowork--which faces major security vulnerabilities.
Jun 27, 2026
A 19-year-old CTO shipped LoginWithChatGPT — a web OAuth wrapper around Codex CLI device codes so indie apps can run prompts on the user's ChatGPT subscription instead of the developer's API key. Here is what it does, whether it violates OpenAI terms, and how it compares to OpenClaw and usemysub.com.
Codex pets sit at the intersection of developer-tool UX and agent observability: tiny sprites that react while Codex runs, waits for you, or errors. The feature ships in the OpenAI Codex desktop experience alongside serious knobs—MCP, skills, browser, computer use—but pets punch above their weight for anyone who keeps long threads warm while multitasking.
This is the follow-on to our Codex pets announcement recap. Where that post focused on what shipped, this one is a field manual: settings, commands, hatch-pet install hygiene, art direction, troubleshooting, and a practical answer to “what are the best Codex pets?” without inventing fake rosters.
Fast paths: hatch-pet on explainx.ai · copy prompts on explainx.ai /generate/codex-pet · browse context on /codex-pets.
Agentic coding UIs share a nasty failure mode: slow, low-salience feedback. A plan can run for minutes; permissions prompts arrive out of order; background work is invisible unless you obsessively tab back. A persistent, lightweight avatar that mirrors state can beat toast spam when it answers: Is Codex still working? Is it blocked on me? Did the last step explode silently?
That is the same design logic behind status LEDs, CI pills in Slack, and Dynamic Island metaphors—just rendered as a Tamagotchi-class sprite. Leadership commentary on social channels argues pets feel “more useful than they sound”; treat that as product intuition, not peer-reviewed UX research.
When pets hurt: teams allergic to motion, accessibility-sensitive setups, or screen-recorded demos where mascots read as unserious. Codex lets you tuck pets away—they should never be mandatory for power users.
OpenAI documents Codex app settings as the home for pet controls under Appearance. The flows you should know:
| Area | Why it matters for pets |
|---|---|
| General | Output visibility and composer ergonomics influence how often you need peripheral status cues. |
| Appearance | Theme, accent, fonts—and Pets: choose built-ins or refresh custom pets from your Codex home directory. |
| Agent configuration | Inherited from shared Codex config; pets do not replace AGENTS.md discipline. |
| Integrations & MCP | MCP servers share config.toml with CLI; unrelated to sprites but part of the same trust surface. |
Command palette: Cmd+K (macOS) / Ctrl+K (Windows) is the fast path for Force Reload Skills, pet toggles mentioned in copy like Wake Pet / Tuck Away Pet, and other Codex commands.
Floating overlay: Pets can stay visible while you context-switch—useful if you batch human reviews while Codex grinds.
Baseline:
/pet — surface or interact with the pet experience from the composer (exact behavior is versioned—verify in-app help after updates).Creator path:
$skill-installer hatch-pet — installs the curated hatch-pet skill from OpenAI’s skills corpus (composer syntax mirrors other Codex skill installers).$hatch-pet … — generation prompts once the skill is live; always reload skills after install.If a teammate shares a screen recording, standardize language: say “force reload skills” before demoing custom sprites so newcomers do not assume flaky loaders.
Custom pets are not undocumented graphics APIs. They ride OpenAI’s agent skills system—the same pattern we explain in What are agent skills?.
Authoritative listing + install: hatch-pet on explainx.ai — source tree: openai/skills → skills/.curated/hatch-pet
Happy path:
pet.json / atlas conventions in the skill README—avoid hand-editing binary layouts unless you enjoy debugging loaders.Supply-chain hygiene: read SKILL.md, pin revisions, and audit any helper skills that call external image APIs.
There is no official leaderboard—OpenAI may rotate built-ins. Instead, evaluate pets on readability and state fidelity:
| Criteria | Great pets | Weak pets |
|---|---|---|
| Silhouette | Reads at small sizes; distinct head/body or glyph | Hair-thin strokes vanish on dark themes |
| State coverage | Idle / busy / blocked / success cues are unmistakable | Only loops idle animation |
| Color discipline | High contrast against Codex chrome; limited palette | Neon gradients that fight UI accents |
| Motion economy | Short loops; no seasickness | Excessive screen shake or particles |
| Personality copy | One-line flavor text that matches team tone | In-jokes that confuse guests on streams |
Documented built-in example: OpenAI’s Codex settings copy cites Seedy with flavor text “A new idea just sprouted from your last prompt.” Use that as a reference for tone—playful, concise, tied to ideation—not as a guarantee every install includes the same roster.
Custom “top picks” by archetype (conceptual):
Use this as a pre-flight list when collaborating with design:
idle_01.png, think_loop.png, etc.) saves engineer time.| Symptom | Likely causes | Mitigations |
|---|---|---|
| Pet never appears | Feature disabled; tucked away; older build | Revisit Appearance → Pets; update Codex; check regional availability notes in official docs |
$skill-installer fails | Skills path misconfigured; offline restrictions | Inspect config.toml MCP/skill entries; verify network; reinstall skill in a clean home dir |
| Custom pet missing after reboot | Assets not in Codex home; loader mismatch | Re-run hatch-pet packaging; confirm paths documented in SKILL.md |
| Animations choppy | Heavy atlas; GPU contention | Reduce frame count; simplify gradients |
| Team bans pets | Compliance / recording policies | Default to tucked state; provide pet-free demo profile |
Always cross-check developers.openai.com/codex/changelog after an update—pet commands have shifted quickly in early 2026 releases.
Pets are fun; skills remain code + assets:
For the broader skills threat model, see Agent skills security.
Do pets improve model quality? No. They change awareness, not reasoning.
Windows vs macOS? Codex targets both; GPU/OS compositor differences mean animation fidelity may not be pixel-identical—test on your fleet.
Can I sell pet packs? Business model is between you and your customers—technically they are assets; legally mind IP and OpenAI mark usage.
Commands, installers, and built-in rosters change with each Codex release. This article reflects May 6, 2026 context—verify Settings, the changelog, and hatch-pet SKILL.md before internal enablement guides go live.