Jinling Yuanbao

1. Meet them

2. Play with them

2′ Click and watch it perform

Pick a pet first, then click a skill below — the one on your desktop will do it right away(Just a try-out — no settings are changed.) If it isn't on screen it will be called back first.

3. Talk to them (voice)

Two ways in:"Voice" in the right-click menu, or ⚙️ on the pet → Skills → Voice (the second one opens the scene panel).

Scene panel

Command list (you don't have to say it exactly — close enough works). Speak English when the interface language is English. Note: poem titles are Chinese, so say "recite" first and the title in Chinese; and for English speech please use a multilingual engine (SenseVoice / whisper.cpp / cloud) — the local Paraformer model only understands Chinese.

3′ Chat (experimental · local small model)

When it doesn't catch your command or a skill, it answers you freely with a small model running on your own machine — not a canned line, but a real reply.

How to turn it on

1. Settings → Voice → Intent understanding → turn on “Chat ability” (local small model)
2. Click “Download model” (one-off: 0.6B ≈ 378MB, 1.7B ≈ 1.06GB) — once it is downloaded the button turns into “✓ Downloaded” and the status line says so, so you will not download it twice.
3. Click "Warm up" to load the model into memory first — replies come faster afterwards. You can skip it, but the very first line waits about 6–8 seconds.

Which size to pick

4. Right-click menu

Clickright-clickon the pet to open it. The menu never covers the pet; drag any empty part of it to move it.

5. Per-pet settings (one set per pet)

Two places: the settings window (the left rail has only Pets / Voice, with 📖 Manual and 🧭 Setup at the top right) and each pet’s own ⚙️ panel (Skills / Voice tabs).

Settings window · Pets page

• Pets: each pet's portrait, scale, on/off desktop; clicking the portrait toggles on/off desktop (a hint shows on hover); the row at the bottom of the portrait holds ⚙️ per-pet settings / ⭐ app face, and pet-package pets also get a 🗑 to uninstall
• Days together: each pet card shows “Together N days · N in a row”. Visit daily — a streak adds bonus bond points (a missed day never subtracts; it just goes back to the base points)
• Display: always on top (keeps it above other windows)
• Language: interface language (Simplified / Traditional Chinese / English), applied instantly; what the pet says and the content are never translated

Settings window · Voice page (optional features)

• Four collapsible groups: Speech recognition / Intent understanding / Pet voice / Data folder — click a title to collapse or expand (collapsed by default)
• Data folder: choose where data lives and move existing content over; a rarely used setting, tucked at the bottom
• 🧭 Setup at the top right reopens the first-run guide (add-ons and data folder)

Config window · If something goes wrong

• When the app freezes, shows an error, or clicks do nothing: hold Ctrl and click the version number in the title bar (the small string starting with v)
• It collects the app version, data folder, pet lineup, memory and CPU usage, and the latest runtime log (takes about 2–5 s)
• Once generated it is saved to a file and selected for you in Explorer — just send that file to me.
• You do not need to understand it — just send the whole file over

Per-pet settings (⚙️ panel)

• Skill tiles: click to preview; the 🔊 at the tile’s bottom-right = whether that skill is read aloud
• When a stroll bumps into the left or right screen edge, it walks one full lap around the screen: clockwise along the screen edges, turning its body 90° at each corner with its feet against the edge (on the top edge it hangs upside down — that is intentional). After one full lap it goes back to ordinary strolling.
• 🧗 Edge stroll: say "climb" / "hug the edge" / "along the edge" / "go around", or click this skill cell — it strolls along the screen edge; if it isn't at the edge yet it hops over first. Click it, say "stop", or move the mouse close and it stops
• Voice: a manually triggered skill; opening it shows the scene panel, which currently has “Say something”

⚙️ panel · Voice tab

• Voice / volume / speed: set individually for this pet
• Walk speed: the row at the top of the Skills tab (Slow / Normal / Fast / Very fast) — this pet’s walking speed; strolling, follow-the-cursor, edge-walking and route walks all use it
• Star trail: the checkbox at the top of the Skills tab — when checked this pet leaves a stardust trail while walking (warm gold for Yuanbao / cool cyan for Erbing, fading after a few seconds); uncheck to turn it off
• Speech bubble: the line above the pet's head. There is an x at its top-right — tap it to close the bubble for this line / for today / forever (tapping x only opens the choices, so you cannot close it forever by accident). Tick the box above to bring it back. This only affects the bubble — not its voice, and nothing else.
• Speak normally (read every bubble): whether everyday bubbles are read aloud
• Speak during voice chat: whether its replies are read aloud after you talk or chat (a separate scene from “Speak normally” above)
• Mute while the computer plays audio: it stays quiet while you watch videos or listen to music (bubble text still shows)
• Which skills speak：The 🔈 icon on each skill tile in the Skills tabis that skill's speak switch — lit = read aloud, greyed out (🔇) = bubble only. By defaultsleep, blink, stop, be quiet and directionsonly show a bubble; the rest are read aloud
• Voice skills group: the last group in the Skills tab (edge stroll / riddle / daily fortune / focus companion / skit theater / come / wake / quiet / stop / left / right) — click one and it does it right away; their 🔊 switches work the same way
• Call the pet (optional):each pet has its own copy — the “Listen for the pet’s name” switch, session mode (one question one answer / continuous) and wake nicknames are all per pet, independent of each other

Choosing a recognition service

By default it uses local recognition sherpa-onnx(Paraformer-zh, Apache-2.0, commercial-friendly) — fully offline, free, no internet, no key; the model (about 240MB) is a one-time download. For cloud recognition (e.g. SiliconFlow) pick another entry in the list and enter your own key. Local whisper.cpp is no longer offered (its models are bigger and slower); existing setups migrate to local recognition automatically.

Two tiers: basic / optional

Only the basic features are on by default — plenty for everyday use; switch optional ones on as needed. Optional features tagged “Experimental” (such as the local small-model chat) are still being explored and vary by machine. The small “Optional” badge at the top right of the Voice item in the rail means exactly that.

Related settings stay together so you never have to hunt around: model and voice-pack downloads sit right where they are used — the offline speech-recognition pack under Speech recognition, local read-aloud voice packs under Pet voice, and the “better understanding” and chat models under Intent understanding. The collapse state of the four voice groups is remembered.

5′ App face (⭐) and pet packages (📦)

Changing the app face: make the app become it

Settings → Pets → hover a pet, then click the ⭐. The taskbar icon, the tray name and every window title switch to its look and name.

It doesn't talk / no sound?

1. Settings → "Pet speech": fill in the speech service (needs nothing by default: Edge online voice, or the local voice once a voice pack is installed)
2. Per-pet settings → Voice tab → switch onSpeak normally; do not turn the volume to 0
3. check whether that skill tile's 🔈 icon is greyed out (grey = bubble only, no sound)
4. it mutes itself while you watch videos ("Mute while the computer plays audio" is on)

It doesn't understand what I say?

1. The first time you need todownload the local recognition modelin Settings (about 240MB, one time)
2. pause after speaking and wait for it to recognise; a noisy room hurts accuracy
3. Commands don't have to be word-perfect, butthe closer you are, the more reliable(e.g. "go right" is more reliable than "go right two steps")

It suddenly disappeared?

It may be "hidden": toggle with a tray left-click, or tray right-click → recall. It may also just have walked to the edge of the screen.

It disappears while I watch videos or play games?

That is our intentional “Hide on full screen”: when a full-screen window (video, game) is detected the pets step aside, and they come back when you leave full screen. If you do not want it, open the config window → Display and turn “Hide on full screen” off.

How do I stop it wandering off / napping?

Per-pet settings → Skills tab, switch off skills like "Stroll" and "Sleep".

Chat (reply with a line) does nothing / answers off-topic?

1. First download the model under Config page → Voice → Intent understanding → Chat (local small model), then warm it up once (the first line always takes a moment).
2. Chat is for small talk: commands and skills it does understand still run as usual and never go to chat.
3. Off-topic answers and repetition are normal for a small local model (it is experimental). If it is not for you, just switch it off — memory is freed right away.