Open-source health agent

Chat with your metabolism.

Innerloop ties your glucose sensor, your Apple Watch and a photo of your lunch into one feedback loop — then puts an agent on top that tells you, in plain language, why your numbers move. Fully open source — build your own.

See how it works View the source
08:125G
‹
Innerloop bot
bot
🩸
How's my swim training coming along? 🏊08:12 ✓✓
Your last 4 sessions:
DateDistanceTime/100m
Jul 121,500 m32:102:08
Jul 051,500 m33:052:12
Jun 281,200 m27:002:15
Jun 211,000 m23:202:20
08:12
Distance +50%, pace 12 s/100m faster — clean upward trend 💪08:12
And how did my heart rate perform?08:13 ✓✓
Heart rate across those swims:
DateMax HRAvg HR
Jul 12178148
Jul 05172149
Jun 28169147
Jun 21164148
08:13
You got faster, but avg HR stayed flat (~148) — more speed at the same effort. Your cardio fitness is improving 🫀📈08:13
📎
Message
➤

The premise

Your body already runs a feedback loop. Your apps just can't see it.

Your CGM knows your glucose. Your watch knows your sleep and training. Your food log knows your carbs. None of them talk to each other — so the one question that matters, what did that actually do to me?, goes unanswered.

SIGNAL 01

Glucose

A continuous trace, every 5 minutes — but stranded in its own app.

SIGNAL 02

Sleep & training

Rest, HR, HRV and workouts — living on your wrist, never cross-referenced.

SIGNAL 03

Food

What you ate and when — tedious to log, disconnected from the rest.

How it works

A photo, a curve, and a conversation.

Log a meal in seconds, watch it hit your glucose, and ask your data anything.

01 — Snap it, watch it land

A photo in the chat. A marker on your curve.

Send a photo to the Telegram bot — a vision model turns it into calories, carbs, protein and fat, and logs it. Seconds later the meal shows up as a marker on your live glucose chart.

The blueberry bowl at 09:40 barely moves the line. The carbonara at 18:40 writes a mountain. Same day, same body — now you can see the difference.

Telegramvision → macroslive meal markers
09:415G
‹
Innerloop bot
bot
🩸
Hey, here's my breakfast 👇09:41 ✓✓
Yogurt bowl with blueberries, granola and peanut butter 09:41 ✓✓
Yogurt bowl with blueberries, granola & peanut butter — logged ✅
kcal 455carbs 38 g protein 19 gfat 21 g
09:41
Low glycemic load — this should barely move your blood sugar. Watch the curve on the right 👉09:42
📎
Message
➤
1 · Log it in the chat
→
21:365G
INNERLOOP
😴
6h32
Sleep
👣
8,956
Steps
❤️
49
Rest HR
🩸
98
mg/dL
BLOOD SUGAR + HEART RATE 24h
🫐 logged · +9 mg/dL
mg/dL bpm
RECENT EVENTS
18:40
🍝 Spaghetti Carbonara
720 kcal · 68 g carbs · +62 mg/dL
15:20
🚴 Outdoor Cycling
42 min · 16.7 km · 23.8 km/h
09:40
🫐 Yogurt bowl with blueberries
455 kcal · 38 g carbs · +9 mg/dL
2 · See it hit the curve
02 — Ask about your sleep

Did the new routine actually work?

Tell the agent what changed. It compares your last week against the two before it — deep sleep, sleep onset, time awake — and shows you the 14-night trend.

Meditation plus a fixed 11 pm bedtime: the data says yes.

bedtime routine14-night trenddeep sleep +50%
07:025G
‹
Innerloop bot
bot
🩸
Morning! I've meditated before bed and fixed my bedtime to 11 pm for the last 7 days 🧘 — can you see a difference in my sleep quality?07:02 ✓✓
Yes — a clear improvement in your deep sleep 📈07:02
This week vs. the two before:
MetricBeforeNow
Deep sleep34 min51 min
Fall asleep23:5123:12
Awake18 min3 min
07:02
The fixed 11 pm bedtime is doing most of the work — consistency beats duration. Keep it up 💪07:03
📎
Message
➤
1 · Ask about the change
→
23:545G
~ Innerloop
↻
SLEEP PHASES · LAST NIGHT
23:1103:3006:58
Core4h 10m
REM1h 34m
Deep51m
Awake3m
14-NIGHT HISTORY
DeepREMCoreAwake
↑ deep sleep +50% since the 11 pm routine
🏠Today
💡Insights
🩸Sugar
🍽️Food
😴Sleep
🏃Workout
💓Heart
2 · See the 14-night trend

The loop

One loop, four signals.

Three of them stream in on their own. One is a photo. Together they answer the question your apps couldn't.

🩸

Glucose

Continuous CGM readings pulled from LibreLinkUp, high/low flags and trend.

every 5 min
😴

Sleep & recovery

Sleep stages, resting heart rate and HRV, streamed from Apple Health.

nightly
🚴

Movement

Steps and workouts — type, distance, pace — overlaid on the glucose curve.

per workout
🍽

Meals

A photo becomes calories, carbs, protein, fat and glycemic load in your diary.

on demand

Under the hood

How it's built.

A single-user system wired from small, boring, reliable parts — an agent at the center, everything reading one database.

LibreLinkUp CGM → n8n · every 5 min
Apple Health → n8n webhook · iOS Shortcut
↓
Neon Postgres — one source of truth
↑ ↓
OpenClaw agent Telegram LLM vision
↓
Vercel PWA dashboard
01

Agent-first

An OpenClaw agent on Telegram handles logging, analysis and proactive nudges, with tools for vision and SQL.

02

Passive ingestion

n8n pulls the CGM every five minutes and receives Apple Health through an iOS Shortcut webhook.

03

One source of truth

Everything lands in Neon Postgres — the dashboard and the agent read the exact same tables.

04

Conversational analytics

Questions become SQL; answers come back as tables and charts inside the chat.

OpenClawTelegram Bot APIn8n Neon PostgresVercel EdgeLLM visionApple Shortcuts

Open source

Build your own.

Innerloop is fully open source. Clone the repo and you get the database schema, the n8n workflows, the agent prompts and the dashboard — a complete blueprint. The setup guide walks you through wiring your own sensor, Telegram bot and dashboard, end to end.

Read the setup guide Browse the repo