Foreword¶

Personal life data is often scattered across multiple apps: sleep tracked on a fitness band, diet tracked in expense apps, exercise tracked in running apps, and schedule tracked in calendars. They don’t communicate with each other, and answering questions like “How much can I still exercise after eating this today” requires manual conversion. The runtime philosophy of DeepSeek Harness (DSH) is “everything is a plugin”. avenyq/dsh-rigui in the community directory is a client plugin that brings life records into a single page card and makes the cards interact with each other. Below is an introduction to what it is, how to install it, and how to use it.

What is this¶

dsh-rigui (Chinese name: “Sundial”) is a DSH client plugin maintained by AvenYQ, categorized under Client, with 0 stars in the community directory. It is positioned as “a playful personal life workbench for DeepSeek Harness”: sleep, diet, exercise, spending, weight, schedule, and to-do items are placed on the same page of cards, and the cards influence each other based on records. The project version in the README is 0.2.0, and the License is MIT.

Core Features¶

Speak and it records. Speak a sentence casually in the top input box. The plugin uses the default model configured in DSH (prioritizing the home’s fast model) to parse it into structured records. It covers: what was eaten (estimates calories for each item, numbers can be edited directly, remembers for next time), spending and income, exercises (what, duration, steps), sleep and wake times, weight, schedule with specific times, and to-dos without specific times. Phrases like “Weight 83.2 a week ago” will be recorded for that specific day. There is a local rule fallback if the AI is unavailable.

Main Card: Weather and Schedule. The main card connects to Open-Meteo for real-time weather (no key required). The top right displays the city, current temperature, and today’s high and low. Click anywhere to enter the weather page: the left side is current weather, and the right side is a 7-day forecast like iOS weather widgets (including precipitation probability, temperature range bars). Click anywhere on the page to flip back. When in a meeting, studying, or exercising, the top right shows how much time is left in the schedule.

Cards are connected. Record a note, and a beam of light connects from the input box to the affected cards, with the cards flashing in sequence. The more you exercise, the more “you can eat today”; unchecking to-dos with amounts automatically records a transaction. “Relationship between them” finds the causal links between sleep, eating, exercise, and spending. Hovering the mouse over it shows a comet-shaped arc connecting the specific numbers in the two cards.

Each card has its own feel. Cards look like ID badges hanging on a line, draggable, and resizable. Flip the card to see how the number was calculated. The sleep card hides a field of stars; the mouse acts like a small light illuminating the nearby area. The exercise bars look like piano keys; touching a day makes the corresponding key jump. The weight card is frosted glass; wipe it a few times to see through. Interactions like check-in, receipts, and undo have their own animations; animations converge if the system “Reduce motion” is enabled.

Goals. The Target button opens “Goals”. It provides six layers of advice based on recorded data (Daily tasks, This month, This year, Vision, Anti-vision, Bottom line), referencing the structure of Dan Koe’s How to fix your entire life in 1 day. Adopting an item throws a dart at the corresponding layer. When all six are lit, the target spins to lock.

Day, Night, and Weather. The entire page is the sky outside: day after sunrise, night after sunset, with a sky switch in the top left of the main card. On sunny days, the card’s projection direction and length follow the sun; on cloudy days, drifting shadows pass by; on rainy or snowy days, rain or snow falls for about 10 seconds when the Sundial is opened. Snow accumulates along the top edge and rounded corners, and thunderstorms flash. Manually toggled switches persist until the next sunrise or sunset. Only slow drifting clouds and fog in the background are rendered by the GPU.

Installation and Enablement¶

The official installation command (web profile) is as follows:

dsh plugin --profile web add github:AvenYQ/dsh-rigui

For the desktop client, replace web with the profile you are using (e.g., desktop). After installation, restart DSH once. A new row “Sundial” will appear in the left sidebar. Click to enter, and enter the city in the top right of the main card to view local weather. Click the greeting to tell it how to address you.

Typical Usage¶

After installing and restarting, describe a day’s record directly in the top input box, for example:

早餐 两个麻团 一杯豆浆 8

After pressing Enter, the calories of the corresponding food, this expense, and today’s timeline will be updated together. Later, entering “Ran 5km tonight” will increase “How much you can eat today”; entering “Meeting next Wednesday at 3 PM” will show a schedule countdown in the top right. To see how numbers interact, hover your mouse over “Relationship between them”.

The project also provides a preview page that doesn’t depend on DSH, convenient for seeing the effect first:

npm install
npm run build   # Generates client.js, index.js, and dist/preview.html

Then open dist/preview.html directly in a browser. Regarding source code structure, src/host.js is the host side (registers /api/rigui route, calls DSH’s model to parse a sentence), and src/plugin.jsx is the browser side (registers sidebar entry and main area page).

Data and Privacy¶

All records are stored in the local browser storage and are not uploaded anywhere. Only the sentence the user inputs is sent to the model configured in DSH for parsing; when checking the weather, only the city name and latitude/longitude are sent to Open-Meteo.

Applicable Scenarios and Notes¶

Suitable for individual users who want to record diet, spending, exercise, and sleep in one entry and hope these data influence each other on the page.

Two points to note: First, the plugin runs with the permissions of the current DSH process. Before installing, you should read src/host.js and src/plugin.jsx to confirm that the content sent to the model and the calling method meet your expectations. Second, the license is MIT, so it can be used and modified freely, but the LICENSE file in the repository takes precedence. The plugin relies on DSH’s default model for natural language parsing. If the AI is unavailable, it degrades to local rule fallback, and parsing accuracy depends on the model used.

Conclusion¶

The value of dsh-rigui lies in gathering scattered life records into one page and explicitly presenting relationships like “How much to eat, exercise, spend today” through card interactions. See the directory page at https://www.skillhub.cn/plugins/AvenYQ/dsh-rigui , source code at https://github.com/AvenYQ/dsh-rigui . The directory site is an independent community site with no official affiliation to DeepSeek / Fudan.