dsh-web-whale-maid (DeepSeek Harness Desktop Pet · Anime Maid Whale)
English | 简体中文
An interactive, anime maid whale desktop pet plugin customized for the DeepSeek Harness (DSH) Web UI.
Features LLM thinking and mood linkage, 30,000 Token satiety pool, persistent speech bubbles, mobile-responsive cupboard modal, 5-hour task awareness ledger, and dual-layer persistence.
✨ Features
- 8 Mood Frame Animations:
idle, hungry, happy, angry, eating, scared, sad, and thinking.- Built from a 1024x2048 HD sprite sheet processed with chroma keying and custom despill algorithms for clean, crisp edges on Retina screens.
- Strict Mood & LLM Linkage:
- The pet's mood and animation only change when an LLM invocation occurs; otherwise, it remains in its current state.
- Thinking Animation: Whenever an LLM call starts (upon feeding or work session completion), the pet immediately transitions into the
thinking animation. When the LLM outputs its response, the pet switches to the result mood and renders its speech bubble.
- Dual-Zone Touch & Persistent Speech Bubble:
- Permanent Bubble: Speech bubbles and moods remain permanently on screen until the next LLM response arrives.
- Clicking Pet Body: Opens the antique wooden maid cupboard modal.
- Clicking Speech Bubble: Immediately dismisses the speech bubble and resets the pet back to standard standby (
idle) mode.
- Sidebar Footer Toggle & Instant Dormant Sleep Mode:
- Dedicated one-click toggle button rendered at the bottom of the DSH sidebar (
sidebar.footer.action beside Settings), fully responsive across expanded and rail modes;
- Features an instant "Put to Sleep" button inside the cupboard header;
- Dormant mode completely unmounts the overlay DOM (zero screen occlusion) and cuts off all LLM session event watchers and Token consumption, with instant millisecond wake-up without page reloads.
- Satiety Reserve Pool (Tokens as Food):
- 30,000 Token maximum capacity.
- Four tiers of rice food items on the shelf:
Grain of Rice (+1 Token)Spoon of Rice (+1,000 Tokens)Bowl of Rice (+3,000 Tokens)Pot of Rice (+10,000 Tokens)
- 5 perception tiers:
Starving! (<2,000), Very Hungry! (2,000-5,000), Not hungry, but wouldn't mind a snack! (5,000-10,000), Satiated & Content (10,000-20,000), and Stuffed! (>20,000).
- Body & Bubble Size Scaling Slider:
- Continuous size scale slider located inside the cupboard drawer (50% to 180% range, 5% step).
- Scales both the whale maid pet and its speech bubble simultaneously in perfect proportion.
- Features real-time percentage display and a quick "Reset" button to 100%, with preferences persisted to disk and localStorage.
- Universal DSH Model Auto-Discovery:
- Automatically parses and synchronizes all providers and models configured in
~/.dsh/settings.yaml (including custom OpenAI-compatible proxies, Ollama, Claude, etc.) alongside DeepSeek official models.
- Automatically identifies the active
agent-default-model as the initial preference, and refreshes models dynamically whenever the cupboard modal opens. Zero machine-specific hardcoding.
- Mobile-Responsive Antique Cupboard Modal:
- Automatically scales down proportionally based on viewport size (
min(92vw/640, 88vh/580)) and centers smoothly on narrow mobile screens, eliminating clipping.
- Instant Close on Feed: Tapping any rice item automatically closes the cupboard modal, allowing the user to watch the pet's thinking animation and response on the main screen right away.
- Smooth mobile momentum scrolling on the speech ledger.
- Speech Ledger & Dual-Layer Persistence:
- Records up to 20 past LLM dialogue entries including moods, timestamps, food/session titles, and tokens used.
- Stored concurrently to disk (
pet_durable_memory.json) and browser localStorage.
- Context Awareness & Strict Prompt Constraints:
- Aggregates tasks completed in DSH within the past 5 hours to generate tailored commentary and avoid repetitive responses.
- Zero Emojis: Strictly forbids all emojis across all prompts, UI elements, and dialogues. Single-line JSON format bounded within 25 Chinese characters.
📦 Installation
Method 1: Ask your DSH Agent (Easiest 🤖)
Send this repository URL directly to your DSH web agent in chat:
"Install this plugin for me: https://github.com/AcidGr/dsh-web-whale-maid"
Your DSH coding agent will automatically configure and mount the plugin.
Method 2: DSH CLI Install (Recommended)
Run the following command in your terminal:
dsh plugin --profile web add dsh-web-whale-maid
(Or install directly from GitHub):
dsh plugin --profile web add github:AcidGr/dsh-web-whale-maid
After installation, refresh your browser.
Method 3: Manual / Offline Install
PROFILE="${DSH_HOME:-$HOME/.dsh}/profiles/web"
mkdir -p "$PROFILE/plugins" "$PROFILE/node_modules/@dsh-profile"
cp -r dsh-web-whale-maid "$PROFILE/plugins/whale-maid"
ln -sfn ../../plugins/whale-maid "$PROFILE/node_modules/@dsh-profile/whale-maid"
# Append to $PROFILE/cordis.patch.yml:
# - insert:
# - id: whale-maid
# name: '@dsh-profile/whale-maid'
🎮 Controls & Interactions
- Drag & Move: Click and drag (or touch and drag on mobile) to position the pet anywhere on the screen. The position persists across reloads.
- Open Cupboard: Click or tap the whale maid's body to open the Maid Cupboard.
- Top shelf: View real-time satiety tokens and percentage.
- Middle shelf: Click food items to feed. The cupboard modal closes automatically.
- Bottom shelf: Configure the active feeding model and browse/clear the speech ledger.
- Task Accompaniment: Whenever a coding session completes, the pet observes the task and comments with appropriate moods and token deductions.
- Dismiss Bubble: Click the speech bubble to close it and immediately return the pet to
idle standby.
📂 Project Structure
dsh-web-whale-maid/
├── assets/
│ ├── pet_assets.json # Base64 assets package (sprites & cupboard background)
│ ├── pet_sprite.png # 8-mood HD sprite sheet (1024x2048)
│ ├── cupboard_bg.png # Antique cupboard background (640x580)
│ └── rice_*.png # Rice food icon assets
├── cordis.patch.yml # DSH Cordis bundle mount specification
├── lib/
│ ├── index.js # Host half: LLM orchestration, mutex queue, disk storage
│ └── client.js # Client half: animations, responsive layout, touch events
├── package.json # npm package metadata and DSH profile configuration
├── LICENSE # MIT License
├── README.md # English documentation
└── README.zh.md # Simplified Chinese documentation
📄 License
This project is open-sourced under the MIT License.
