NOOP AI

Your WHOOP data, on your own devices, with a coach that actually remembers you.

---

What you get

⌚ Your WHOOP, without WHOOP	Pairs straight with a WHOOP 4.0 or 5.0/MG over Bluetooth. No WHOOP account, no subscription, nothing in the middle.
📱 iPhone and Mac	Runs on iOS 17+ and macOS 13+, from the same codebase.
🤖 A coach with a memory	22 tools it uses to look things up mid-answer, a goal, a plan you agreed to, and facts it remembers across months — not a chatbot bolted onto a dashboard.
📈 Deep health analysis	Recovery, strain, sleep, HRV, readiness, training load and n-of-1 correlations — all computed on your device, from published methods.
🔘 A home-screen widget, and Siri	The NOOP Rings widget (Charge alone, or Charge + Effort + Rest) at a glance, plus a "How's my recovery?" Siri/Shortcuts intent that answers without opening the app.
🔒 No server, no account	Everything lives in a database on your machine. The coach is the one thing that opens a socket, only with your own API key, only when you ask it something.
🍎 Apple-only, diverging freely	Forked from ryanbr/noop — every protocol decoder and analytics formula started there — but this fork dropped Android and no longer tracks upstream commit-for-commit.

New here? The coach is what makes this fork different — jump to The coach. Everything else is upstream NOOP, which is excellent and deserves the credit.

---

Contents

• The coach — the reason this fork exists
• What NOOP itself does · What leaves your device
• Quickstart · What it looks like
• About the fork · Docs
• Attribution · Disclaimer · License

---

The coach 🤖

Upstream ships a chat-with-your-own-API-key coach that receives one pre-baked block of text per message. This fork turns that into something with a goal, a plan, a memory, and hands to go and fetch its own data.

🔧 It looks things up while it answers

Instead of a fixed summary, the coach reaches for 22 tools mid-sentence. Crucially, get_readiness and get_charge_drivers read from the exact same engines the Today screen uses, so its verdict can never contradict the number you're looking at.

• 🎯 get_readiness — the same push / maintain / rest verdict Today shows
• 🔬 get_charge_drivers — why today's Charge is what it is, term by term, never an invented reason
• 📈 plot_metric — draws a real chart inline in the chat
• 📝 propose_plan — suggests a session; it cannot book one
• ☕ log_caffeine / log_journal / log_lab_marker — writes to your real app data

That last group is the fun one: "just had a double espresso" becomes a real entry in the Caffeine card. "my Vitamin D came back at 38" becomes a Lab Book marker. Same data, logged by talking.

🧠 It has a real memory

Most AI chat features either forget everything or dump everything into every prompt. This one works more like a person's:

• Facts have shape. Each carries a category (goal, injury, preference, physiology, schedule) and an importance. Pinned facts — a serious injury, a hard constraint — ride along on every reply; the rest surface only when relevant to what you just asked.
• It corrects itself. Tell it your knee is fine now and the stale fact is rewritten, not stacked next to a contradiction. Near-duplicates merge instead of eating the memory budget.
• It remembers conversations — including when. Ask "what did I ask you yesterday?" and it searches by day, quoting your own questions back.
• It tidies up cheaply. Leaving a chat hands it to a small, cheap model (Haiku / gpt-4o-mini / Flash-Lite — you pick) for a one-line summary. Your expensive model never pays for housekeeping.

Every fact is visible and editable in Settings. Nothing is remembered that you can't see or delete.

🎯 It knows your goal, and stays honest about it

Set a target — a run, a consistency streak, sleep, weight, or free-text — and the coach stops improvising. Up to five goals can run at once, weighed against each other.

• Two safety checks, neither of which ever blocks you. Is the pace aggressive? Measured as a percentage of your own body weight or volume per week, so the same 20 kg cut reads correctly at 60 kg and at 160 kg — an aggressive pace asks for a one-line reason and remembers it. Is it realistic? Checked against your on-device VO₂max estimate — and for weight goals, always reported honestly as "no data to judge that on", because there isn't any.
• It never plans your diet. Training is the only lever it has, and it says so.
• No invented percentages. Progress is a real measurement against a real baseline, or nothing.
• A goal whose date passes gets a look-back — once, asking what you want to do with it, rather than a verdict on a number it hasn't checked.

📅 It proposes; you decide

The coach can suggest a session. It cannot schedule one. Every suggestion waits as a proposal until you accept, decline, reschedule or swap it.

Swapping shows the consequence before you decide, computed from your own history rather than a generic table: "CrossFit at 10:00 instead of Zone 2 — about 18 points and 2 recovery days instead of 6 and one. Tomorrow's projection drops from ~62 to ~45." Skipping is one tap and a reason; pain and illness are read back with context, not judgment, and a few skips never get a sport permanently written off.

📬 It decides what reaches you, and how

Not every message from the coach deserves the same treatment — a suggested session needs a decision; "your recovery is low" doesn't. The coach now picks a category, a priority, and how long something stays relevant, and the app renders each one differently instead of stuffing everything into one "Today's session" card:

• Needs a decision — a proposed session, with Accept / Change / Decline right there.
• Just a hint — a proactive nudge (a body signal worth knowing, a small win) that used to live only in chat, now also lands in the bell so it's visible without opening a conversation. Dismiss it or leave it for later; there's nothing to accept.
• Status & reminders — read-only history: what a decided proposal turned into, a goal deadline, a restorable card.

The bell — previously Classic Today only — is now in both Today screens, reading the same inbox, so switching between them never hides what's waiting for you.

🎭 It has a face and a name

Svea or Marv — two ready-made identities, each with its own name and avatar — or make your own: any name, a symbol, or a real photo via the system picker. A photo never leaves the device or gets sent to your provider; it's rendered locally everywhere the coach shows up. Identity is who's talking — separate from coaching style (Guardian / Friend / Commander), which only changes tone, never the methodology or the "I'm not a doctor" guardrails.

💬 The chat itself

Streams everywhere	Token-by-token on every provider, with tool calls running inline — including local models, where it matters most.
Says when it spoke first	A daily brief or an unprompted nudge is labelled as such, and badged, so it never reads as a reply to a question you've forgotten asking.
Recovers from failures	Offline says offline; a rejected key offers the key screen; a rate limit counts down from the provider's own Retry-After.
Searchable, pinnable history	Full-text search across threads, pin the ones worth keeping, share any as Markdown.
A closer look on demand	Optionally set a heavier model for one question at a time — "Look at this more closely" on a reply you've already read. Off unless you configure it.
In-chat charts	Native trend charts drawn in the conversation, and they survive a relaunch.
"Ask coach" on any card	A sparkle on Today's rings and every card — a short, careful read of one number without leaving Today.
A name and a face	See 🎭 It has a face and a name above.
Honest about cost	Token counts and cache behaviour after every question, on every provider.
Bring almost any model	Anthropic, OpenAI, Google Gemini, OpenRouter, or any OpenAI-compatible endpoint — including a local Ollama or LM Studio server.
Replies in your language	The coach is told the app's current language and answers in it — chat, check-ins and briefs alike, not just the UI chrome.
8 languages in the UI	English, Deutsch, Español, Français, Italiano, Português, Русский, 中文(简/繁). German/Spanish/French are reviewed; the rest are machine-translated and would benefit from a native pass — see docs/COACH.md.

📖 The deep version → docs/COACH.md: every tool's schema, the safety gates, the plan book's state machine, the memory ranking, provider support, and the file map.

---

What NOOP itself does ⌚

Everything in this section is upstream ryanbr/noop, inherited unchanged — credit belongs there and in docs/FEATURES.md.

• Pairs directly with a WHOOP 4.0 or 5.0/MG strap over Bluetooth Low Energy. WHOOP straps don't appear in Settings → Bluetooth; NOOP finds them on their own advertising profile.
• Computes its own scores on-device, from published methods: Charge (recovery), Effort (strain) and Rest (sleep quality) — an energy economy you wake with, spend, and rebuild overnight — plus HRV, resting heart rate, SpO₂, respiration and skin temperature. Honest approximations, explicitly not WHOOP's own scores.
• Everything lives in an on-device SQLite database. Import a WHOOP or Apple Health export for instant history, or just wear the strap and let it build.
• Automatic Apple Health sync (HealthKit background delivery), so the coach always reasons over fresh data.

What leaves your device 🔒

Worth being precise about, because "AI" and "private" rarely share a sentence:

• The app itself is fully offline. Strap data, database, scores, goals, plan, memory and chat history are all local. There is no NOOP server and no account.
• Only the coach opens a socket — only when you send a message, only to your own provider with your own key, and only if you've turned data access on. Turn it off and the coach still works; it just doesn't see your numbers.
• Data access isn't one switch — it's seven. Core biometrics, workouts, planning, stress, logging, memory and personal patterns are each granted independently (Settings → Privacy & data → Data access), so you can hand the coach your workouts without your sleep, say, rather than all-or-nothing.
• Summaries, never raw signal. Derived daily numbers and short text — never raw R-R streams or sensor buffers.
• You can go fully local. Point the Custom provider at Ollama or LM Studio and nothing leaves your network at all.

More: docs/PRIVACY_SECURITY.md.

Quickstart 🚀

You'll need a Mac with Xcode 26+ and xcodegen (brew install xcodegen). For the iOS app you also need a physical iPhone — Bluetooth and HealthKit don't exist in the Simulator.

git clone https://github.com/DX23876/noop.git NOOP-AI
cd NOOP-AI
xcodegen generate     # project.yml is the source of truth; the .xcodeproj is generated
open Strand.xcodeproj

iPhone — pick the NOOPiOS scheme → select your iPhone → Signing & Capabilities → set your Team → ⌘R. On the phone, trust the certificate under Settings → General → VPN & Device Management. A free Apple ID works, with two small trade-offs (the Watch app/widget step back, and the build expires after 7 days) — 👉 docs/DETAILS.md.

Mac — pick the Strand scheme and ⌘R.

Then: pair your strap → grant Apple Health access → open Coach → paste your API key → optionally give your coach a name, a face and a goal.

What it looks like 📸

Svea	Marv
Pick a ready-made coach, or make your own — any name, symbol or photo.

▶️ Short demo video: marketing/NOOP-demo.mp4.

About the fork 🍴

NOOP AI is a personal fork of ryanbr/noop — not a competitor, and not a rebrand that hides where it came from. Every protocol decoder, analytics formula and pixel of the design system comes from upstream; this fork adds a much bigger coach on top, in its own files, without rewriting upstream logic. Apple platforms (iOS + macOS) are what it builds and tests — Android is dropped as a target, and this fork no longer necessarily tracks upstream; it diverges freely now.

Just want a great WHOOP app? Use ryanbr/noop — macOS, Android and iOS, actively maintained, and the right answer for most people. This fork is for the narrow case where you want the coach pushed further than a cross-platform project could reasonably justify.

👉 The full rationale: docs/DETAILS.md.

Docs 📚

• docs/COACH.md — the coach in full: tools, goal gates, plan book, memory, providers.
• docs/DETAILS.md — fork rationale, the complete tool table, architecture, signing fine print, where things stand with upstream, and the docs index.
• docs/PRIVACY_SECURITY.md — the data posture in detail.

---

Attribution 🙏

NOOP AI is a fork of NOOP by ryanbr — please treat that repository as the canonical project, not this fork. NOOP itself stands on community protocol-documentation work: johnmiddleton12/my-whoop (WHOOP 4.0 BLE protocol), b-nnett/goose (WHOOP 5.0/MG BLE protocol), groue/GRDB.swift (SQLite persistence), and weichsel/ZIPFoundation (export unzipping).

NOOP contains no WHOOP proprietary code, firmware, logos, or assets. Full detail in docs/DETAILS.md and ATTRIBUTION.md.

Disclaimer ⚠️

NOOP AI is an independent, unofficial, non-commercial interoperability project. It is not affiliated with, endorsed by, or connected to WHOOP, Inc. All references to "WHOOP" are nominative.

NOOP is not a medical device. Heart rate, HRV, recovery, strain, sleep stages, SpO₂, respiratory rate and skin temperature are approximations from published methods — not clinically validated, not medical advice. The AI coach is not a doctor and must not be used to diagnose or treat. Consult a qualified professional. Provided as-is, with no warranty, for personal and educational use. See DISCLAIMER.md.

License 📄

Source-available under the PolyForm Noncommercial License 1.0.0: free for personal and other non-commercial use — read it, run it, fork it. Commercial use is not granted. This fork keeps the upstream LICENSE and Copyright 2026 NoopApp notice intact, per NOOP's mirroring terms; bundled dependencies keep their own licenses (see NOTICE).