MIT Python plugin (plus a SwiftUI iPhone app) that makes DeepSeek Harness a local-first health coach — kernel-computed metrics, generative todos, three recommendations a day, follow-along video cutting and an insight scan every three days, with all health data kept on your own Mac.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-16
License
MIT
01What can it help you accomplish?
Run a full health loop inside a DeepSeek Harness session: read Apple Health and WHOOP data on your own Mac, surface the one number that matters today, plan todos, remind you on time and record whether it worked
Four board pages (now / rest / sport / stress), each with one main card for the most problematic number, at most 3 upcoming todos with real times, lock-screen pushes, and a completion record the next scan can judge
People who want to manage sleep, training, rehab and stress themselves, with the data staying on a Mac they control, and who would rather be pushed than shown another dashboard
Turn a trainer's demonstration clip from Xiaohongshu or Bilibili into a follow-along video you can actually train with
A plan card first (which movements, which original segment, seconds per side, rounds, expected runtime) and only after your confirmation the cut: the trainer's explanation segment, then a seamless original-speed single-person loop with per-second counting and credits, checked item by item before delivery
People doing stretching, strength, rehab or relaxation work at home who want a video to follow instead of a raw clip
Get a judgement about yourself every few days instead of another data dashboard: which nudges never work, where the loop breaks, what co-occurs across signals, and what you have not noticed
At most 3 points, each one conclusion plus a sentence or two of support and an optional small chart, then a card to pick what to act on first; when nothing qualifies it stays quiet
Users who want to be told what to change, with every number taken from the kernel instead of model guesswork
02How to install into DeepSeek Harness
Prerequisites
- Runs on your own Mac — the README states the data stays on your computer
- The runtime needs only the built-in Python 3.9+ (the plugin's environment uses the Python 3.12 that DeepSeek Harness ships) and no third-party libraries
- Reading DSH conversation history needs either `zstandard` or the decompression from the Node runtime DeepSeek Harness ships
- Pairing needs the phone and the Mac on the same Wi-Fi; Tailscale covers being away from home
- Follow-along video cutting needs pivis-follow installed on the same machine (it is not in this repo) and `follow_enabled` switched on in the config
Installation steps
- 01
In a new DeepSeek Harness conversation, ask it to install P-Health: first run `dsh plugin --profile phealth add github:PivisYYY/pivis-health`, then `~/.dsh/profiles/phealth/node_modules/dsh-plugin-phealth/bin/phealth setup`, and finally hand you the QR code that appears
$ dsh plugin --profile phealth add github:PivisYYY/pivis-health
- 02
Or register `https://raw.githubusercontent.com/PivisYYY/pivis-health/main/dsh/market-index.json` under DSH's Plugins › Discover › Add index source, install 「P-Health 私人健康管家」 and tell the P-Health assistant to start the install
- 03
Install the iPhone app from the TestFlight public link (the README leaves the link until external TestFlight review passes)
- 04
Open the app, tap 扫码配对 and scan the QR code shown on the Mac
Verify the integration
- `phealth setup` prints a pairing QR code — scanning it in the app is what proves the phone reached the local server
- `python3 -m pivis_health.cli doctor` self-checks that the store is readable and reports how fresh every data source is
- `python3 -m pivis_health.cli whoop-status` shows whether the WHOOP snapshot is fresh without going online
03DSH integration and capability boundaries
DSH plugin — `dsh plugin --profile phealth add github:PivisYYY/pivis-health` installs the `dsh-plugin-phealth` package into its own `phealth` profile, `phealth setup` starts the local server and prints a pairing QR code, and the P-Health agent runs on DSH, fed by a workspace instruction file and the repo's own `./pivis-health` CLI
Generative page orchestration (four boards, main card, open canvas)
The evidence the kernel computes — metric values, which charts can be drawn, projects and todos→A page JSON that is only published after the server validates it; P-Health picks the main-card rendering and assembles canvas cards from 55 parts while the kernel computes the numbers
a page that fails validation is sent back to P-Health to be rewrittena requested change is previewed in the chat and applied only after you confirm, then the page swaps itself a minute or two laterGenerative todos and long-running habit projects
Long-running project templates (a 6-week CBT-I course, knee rehab, a weight-loss route) and 13 todo forms→Today's todos generated per project, proactive recommendations three times a day, and a review proposal card per active project on Sunday morning
a tap answers within a second; offline entries are queued on the phone and replayed once, so a duplicate submission counts oncerecommendations are pushed at 08:00, 13:00 and 18:00, and the morning one also reviews how yesterday wentFollow-along video cutting through the local pivis-follow tool
A trainer demonstration clip you share in the chat→A read-the-clip verdict first, then a plan card, and only after your confirmation the cut video: explanation segment, seamless single-person loop at original speed, per-second counting and credits
the server shells out to pivis-follow on the same machine through `./pivis-health follow …`the resulting video stays on your machine and is never uploaded to a public siteInsight scans with feedback learning
Three days of kernel-computed behaviour: reminders that ring without action, where each managed problem sits in the loop, and predefined cross-signal pairs→At most 3 points that land on an action, with numbers drawn only from the candidate set, plus a card to pick which one to act on; adoption and rejection feed the next scan
the push entry point is off by default — try a few rounds and read the results before enabling itthe server validates every line: numbers must be in the candidate set, no causal claims, no diagnosis, no lecturingNotification and alarm delivery with fallbacks
Todos with real times, chat turns, new questions, and timed forms such as a nap or a foot soak→One push per event (N minutes early, at the time and 15 minutes past), a phone-side system alarm for timed forms that rings even on silent, and an unread dot on the matching entry
the server re-checks time-bearing todos every 20 seconds without calling the modelfailed pushes are retried, replayed within 15 minutes after a restart, and backed by a per-minute task plus an hourly self-checkP-Health deliberately creates no scheduled task of its own — anything that must fire on time becomes a todo card with a time
04Who is it for? When not to use it?
Good for
- People who want to manage sleep, training, rehab and stress themselves, with the data staying on a Mac they control, and who would rather be pushed than shown another dashboard
- People doing stretching, strength, rehab or relaxation work at home who want a video to follow instead of a raw clip
- Users who want to be told what to change, with every number taken from the kernel instead of model guesswork
Not for
- Local-first: everything runs on your own Mac and the data stays there; location and Bluetooth scene data are computed on the home server and only a 在家 / 外出 / 外宿 label is sent to the model; paths, addresses and credentials sit in `config.local.json`, which is not committed.
- Not a medical device: the README rules out diagnosis, treatment, medication and emergency monitoring, and states it cannot replace a doctor. The agent is told to give professional judgement and red-flag triage while avoiding disease names and prescriptions.
- The product is an iPhone app plus a local server: the app must be installed (TestFlight — the public link is left for after external review passes), paired by QR code, with phone and Mac on the same Wi-Fi or both on Tailscale.
05Compatibility, maintenance and safety notes
- Local-first: everything runs on your own Mac and the data stays there; location and Bluetooth scene data are computed on the home server and only a 在家 / 外出 / 外宿 label is sent to the model; paths, addresses and credentials sit in `config.local.json`, which is not committed.
- Not a medical device: the README rules out diagnosis, treatment, medication and emergency monitoring, and states it cannot replace a doctor. The agent is told to give professional judgement and red-flag triage while avoiding disease names and prescriptions.
- The product is an iPhone app plus a local server: the app must be installed (TestFlight — the public link is left for after external review passes), paired by QR code, with phone and Mac on the same Wi-Fi or both on Tailscale.
- Follow-along video cutting depends on pivis-follow being installed on the same machine (it is not in this repository) with `follow_enabled` switched on in the config; without it practice habits cannot get a clip.
MIT · no release yet · local-first health kernel + DSH plugin, build 61
06Frequently asked questions
How do I install pivis-health into DeepSeek Harness?
Open a new DSH conversation and ask for the install, or run it yourself: `dsh plugin --profile phealth add github:PivisYYY/pivis-health`, then `~/.dsh/profiles/phealth/node_modules/dsh-plugin-phealth/bin/phealth setup`, and scan the QR code it prints with the iPhone app. The alternative is DSH's Plugins › Discover › Add index source pointed at the repo's `dsh/market-index.json`.
What does the plugin actually add to a DSH session?
A P-Health agent that runs on DSH: every turn it sees the full brief the server wrote into the workspace instruction file, pulls numbers from the kernel through `./pivis-health`, and returns page JSON that the server validates before publishing. It answers in the main chat plus the rest, sport and stress board chats, and can also be pulled into a DS Bot team group.
Is the DSH plugin the same thing as the Python package?
The plugin is the packaging: `dsh-plugin-phealth` brings the server, the DSH-specific configuration and the background tasks onto your Mac in one command, and `phealth setup` starts it and prints the pairing code. The kernel and the CLI underneath are the repo's own `pivis_health` module — `python3 -m pivis_health.cli doctor`, `page-v3`, `brief` and the rest drive the same code from a terminal.
Where does my health data go, and what does the model see?
It stays on your Mac. Location and Bluetooth scene events are reduced to a 在家 / 外出 / 外宿 label before the agent sees them, and paths, addresses and credentials live in `config.local.json` which never enters the repo. Follow-along videos are served only from your own server.
What are the limits I should know before installing?
It is a single-user personal tool, not a medical device — the README states it is not for diagnosis, treatment, medication or emergency monitoring and cannot replace a doctor. Cutting follow-along videos additionally needs pivis-follow installed locally with `follow_enabled` on, and reading DSH conversation history needs `zstandard` or DSH's bundled Node.
07Related DSH workflows
dsh-univer-office
by dream-num
Give DeepSeek Harness a real office environment. Univer Office Plugin brings spreadsheets, docs, slides, canvases, relational tables, and more into one runtime — with connected data, validation, versioned changes, and isolated worktrees for multi-agent collaboration.
clearai-dsh
by clearailhc
Brings an evidence-checked epistemic loop to DSH: research sessions settle into a browsable domain ontology with evidence chains and support levels; contradictory readings are surfaced for human adjudication.
dsh-fish-memory
by yuuyuko-uu
Stores time-stamped, decaying per-session memories, recalls relevant items, and optionally distills imported local archives through the configured DSH model.
dsh-trading
by zhu1090093659
Agent-native trading terminal built on DeepSeek Harness. Crypto, US, CN and HK in one three-column GUI, 19+ hot-swappable connectors, dry-run by default with human approval on every live order. BYOK, no data redistribution.
08Data and sources
发布成 DeepSeek Harness 插件:一条命令在自己的 Mac 上装好服务器、DSH 专用配置和后台任务,App 扫码配对
- **P-Health Agent**(跑在 DSH 上):页面上写什么、主卡画什么图、待办怎么排、对话怎么回,都由它根据真实数据决定。
它写在工作区的指令文件里,DSH 每轮自动带给 P-Health,数据变了才更新。
- 它也能被拉进 DS Bot 团队群,和别的 Bot 一起讨论;群里定下的事当场落到计划里,App 这边接着跟进。
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-16. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.
