Dashboard
window1 week03-31 → 04-07
visible7,906records
storage16,180last 30 days
stream● livehealthkit · LAN
/ weekly distance /
+12.4 km vs last week
/ this week /
avg 6.1h · target 7.0h
/ overnight hrv vs mg /
triple-shot days vs baseline
Runs the LLM. Schedules every check. Owns the encrypted store.
A single brain that reasons across every surface you already use.
50+ HealthKit signals → server. Watch & phone complications.
Telegram · Feishu · WeChat. Ask, nudge, generate pages.
self-hosted · reads every wearable signal · nothing leaves your machine.
one glance at the room and you know your state — no charts, no scrolling.
every spike, every dip, every overnight HRV sample — analyzed the moment it lands.
Telegram · Feishu · WeChat. Ask anything. Get nudged. Markdown back.
describe it. the agent builds it inside your iOS app, tailored to your data.
a local web console for every metric, every reasoning step, every prompt.
complication, watch face, live state. glance, then move on.
server, app, conversation — a single brain across surfaces you already use.
no cloud. no analytics. no server we own. line-by-line auditable on GitHub.
pick a messenger, boot the server, install the app. ten minutes.
open source · PolyForm Noncommercial · v1.4
HiMe is a self-hosted LLM agent for your body. It reads every wearable signal — heart rate, HRV, SpO₂, sleep, steps, workouts — and reasons about them 24/7. It schedules its own checks, builds personal reports, and pings you proactively. Nothing leaves your house.
Think of it as a junior cardiologist + sleep coach + running analyst, on permanent call, that only talks to you.
your machine, your keys, your LLM.
doesn't wait to be asked.
open source on GitHub.
The agent tags every hour of data with one of six "moods" — relaxed, recovering, active, sleepy, alert, curious. Your HiMeow takes on that mood in the home view, on the watch face, and in chat. It's an ambient health glance: one look, you know.
Under the hood: a small classifier on top of the LLM's hourly summary. You can tune the rules with a markdown skill.
HiMe doesn't just respond to questions. It runs a continuous schedule across every 24 hours and acts on triggered conditions:
Every trigger is a markdown skill. Add your own; the agent picks it up on next reload.
Three gateways supported out of the box:
weekly physical reports, charts inline.
exercise analysis, mile-splits, recovery.
sleep tracking, gentle nudges.
Slash-commands: /evidence dumps raw numbers, /export pdf emails a weekly, /skill name loads a markdown brain.
HiMe's generative-pages engine turns a one-line prompt into a working, data-bound iOS view inside the companion app. The agent picks the right metrics, lays them out, ships them to your phone. No code. No App Store update.
Some examples:
Browse every metric, every reasoning step, every prompt the agent ran. Edit skills as markdown. Audit the data window. Restart, pause, schedule.
streaming records, per-second.
markdown brains you can edit.
papers, baselines, your priors.
The HiMeow lives on your watch face as a complication. Modular Compact, App, and Infograph styles. Glance: the agent's read of your body, current state, top four metrics. One tap opens the app.
The server runs on your machine and owns the data. The iOS app is the wearable's bridge to that server. The chat bot is where you talk to the agent. They share state through a single encrypted store.
Server ↔ App ↔ Chat — three sides of the same brain.
The server is software you install. The wearable data is pushed over LAN. The LLM is yours.
Health records and skills are encrypted at rest with a key only on your machine.
Every line on GitHub under PolyForm Noncommercial. We can't add trackers.
HiMe is research-grade software for personal use. Not a medical device. Not a diagnosis.
Telegram, Feishu, or WeChat. Save the bot token.
git clone, cd, ./setup.sh. That's it.
App Store → Settings → Server URL. Send the bot a message. It replies.
Full install guide on the GitHub repo. PRs welcome.