Internal Technical Brief • March 2026

OneSync

The biological intelligence layer for the agentic economy

Agent OS Architecture • Full Vision • MVP Scope • Possibilities


OpenFangPaperclipOpenVikingCoPawPi MonoAgency-Agentscontext-hub+5 more

Architecture distilled from 12 open-source agent projects into OneSync's health agent OS

The Problem We're Solving

Hundreds of millions of people wear devices that measure them all night, open an app in the morning, glance at a score, and make the exact same decisions they were going to make anyway.

Meanwhile, every AI agent (Lindy, Manus, OpenClaw) can schedule your meetings and draft your emails. None know when you're cognitively depleted. They'll schedule your hardest meeting when your nervous system is wrecked.

We built an entire industry around collecting signals from inside your body — and left people completely alone with it.

Health Apps Are Passive

Dashboard retention at Day 30: 3-12% across the industry. Headspace 7.65%, Calm 8.34%. People stop opening the app. Statista, SensorTower app retention benchmarks 2025

Wearables Collect, Nobody Acts

~500M smartwatches worldwide collecting HR, HRV, sleep every night. The data sits in HealthKit / Health Connect. No system reads it and proactively does something useful. Counterpoint Research, Global Smartwatch Shipments 2025

Agents Are Blind to the Body

YC 2026: ~50% of latest batch is AI agents. NVIDIA GTC 2026: autonomous agents declared the next infra layer. All operate without the most important context: the human's cognitive state. YC RFS 2026, NVIDIA GTC 2026 keynote

What It Actually Looks Like

Real message formats. Telegram delivery. Feedback buttons for learning.

Morning Cognitive Brief (daily at wake time)
You slept 6.8h (good!) but your HRV is still recovering from yesterday.

CRS at 58 — protect your 10-11am slot for deep focus.

→ Move anything optional to after lunch.

Got it More details Mute today
Stress Alert (when HRV drops + sustained)
Your HRV dropped 22% in the last 30 minutes.

This usually happens during back-to-back meetings without a break.

→ Take 3 slow breaths. 4s in, 7s out.

Helpful Not helpful Too frequent

Morning Brief

Every morning at wake time. Uses overnight data (most reliable). The hero feature.

Stress Alerts

Only when confidence ≥ 0.60 AND 2h cooldown clear AND not during sleep. Max 3/day.

Conversational Chat

User messages bot anytime. Full context: CRS + health + memory. All 8 tools available.

🎯

The Full Vision: Your Personal Agent OS

Beyond health messages — you talk to the agent to give it tasks, ask it to schedule meetings around your CRS, teach it new skills ("when I say 'board prep', pull my stats and draft talking points"), and automate anything in your life. It connects to your calendar, email, Slack, task manager — and knows your body while doing it. You teach it once, it remembers forever. Over time, it becomes a full personal operating system that acts on your behalf — with the one signal no other agent has: your biology.

Schedule around my energy Teach a new skill Prep me for standup Decline low-priority meetings Wind down routine

Where We Stand

No one combines deep body awareness + proactive external messaging + device-agnostic + cross-platform.

BodyProactiveExternalDevicesPlatform
OneSyncDeepTelegramYesAnyBoth
WHOOPDeepIn-appNoWHOOPBoth
OuraDeepNoNoOuraBoth
Nori (YC)Agg.LimitedNoMultiiOS
LindyNoneYesYesN/AWeb

WHOOP Coach: confirmed in-app only, $199-359/yr, GPT-4 powered. Nori: iOS only, no Android, data sync issues reported by users. Apple Health+: scaled back per Bloomberg Feb 2026.

Our 5 Differentiators

1. Device-agnostic — any HealthKit / Health Connect wearable
2. External messaging — Telegram (104M users in India alone)
3. Price — free / ₹399/mo vs WHOOP $199-359/yr
4. CRS is transparent — algorithm published, open-source after Phase G. No competitor does this.
5. Cross-platform from Day 1 — Android + iOS simultaneously

The Window

Apple Health+ scaled back. WHOOP is device-locked. Nori is iOS-only. The combination of body awareness + proactive agency + open ecosystem is unoccupied.

What We Validate Before Writing Code

We don't build until we know people care. Phase A is a Wizard of Oz test — manual CRS messages on Telegram, no app, no code.

The Test (Phase A)

5-7 people, 5-7 days

Founder + teammates with Apple Watch / Pixel Watch / Fitbit. Manually compute CRS from wearable data each morning. Send personalized morning brief via Telegram. Track: do they read it? Do they reply? Do they find it useful?

Gate: 3+ people say morning briefs are useful

If nobody cares about "Your CRS is 58, protect your 10am slot" — we stop and rethink the product, not the code.

What We Learn

Message format that works. Timing that works. Whether people even want proactive health messages. Whether CRS zones feel meaningful. All before a single line of code.

What's Proven vs What's Assumed

ClaimStatus
HRV drops under acute stressPROVEN
Sleep quality predicts next-day cognitionPROVEN
Circadian rhythm affects performancePROVEN
Wrist HRV correlates with ECG HRVPROVEN
Our CRS weights (35/25/25/15) are rightASSUMED
CRS zone thresholds (80/50) are meaningfulASSUMED
Proactive messages improve outcomesASSUMED
Rules pre-filter saves 60-80% of AI callsASSUMED
Personal baselines stabilize in 7 daysPARTIAL

The science is solid. Our specific formula is a hypothesis. The WoO test validates the product, not the algorithm.

The Full Vision

OneSync is not a health app. It's a personal cognitive operating system — an AI agent that understands your biology, your work, and your goals, and proactively optimizes your day before you have to think about it.

Layer 1: Body Intelligence

Reads HRV, HR, sleep, activity from any wearable. Computes CRS. Detects stress. All on-phone, offline.

Layer 2: Cognitive Agent

Claude-powered agent with personality, memory, learning. Proactively messages you. Learns your patterns over months.

Layer 3: World Integration

Calendar, email, Slack, tasks. Agent that knows your body AND your world. Schedules recovery before you crash.

CRS
Cognitive Readiness Score
SAFTE-FAST grounded
8→50+
Tools: MVP to full
Health → Workspace → Life
4
Intelligence Levels
Reactive → Autonomous
12
OSS repos synthesized
into our Agent OS

How It Works — The Simple Version

The entire product in 5 steps. No jargon.

1. Your watch measures your body all night

Heart rate, heart rate variability (HRV), sleep stages, steps. Data sits in HealthKit (iPhone) or Health Connect (Android). Today, nobody does anything useful with it.

🧠

2. The app computes your Cognitive Readiness Score (CRS)

A 0-100 score: how sharp are you RIGHT NOW? Based on how you slept, your nervous system state (HRV), your circadian rhythm, and how active you've been. Computed on your phone — works offline, no server needed.

🤖

3. If something matters, the AI agent crafts a message

A rules engine checks every 15 minutes. 60-80% of the time, everything's fine — no AI call, $0 cost. When your CRS drops or stress is detected, Claude Haiku generates a personalized message with one specific action you can take.

💬

4. You get a Telegram message — before you even notice something's off

Morning brief at wake time. Stress alert when your body signals trouble. You can also message the bot anytime: "Why am I so tired?" and get an answer grounded in YOUR data, not generic advice.

📈

5. You tap Helpful / Not Helpful — the agent learns

Every interaction tunes the system. Over weeks, the agent knows YOUR stress patterns, YOUR recovery strategies, YOUR communication style. The longer you use it, the smarter it gets. Switching cost becomes total.

Apple Watch / Pixel Watch / Fitbit

Best data: true HRV (RMSSD), 4-stage sleep, beat-to-beat heart data. Full CRS confidence.

Samsung Galaxy Watch

Samsung deliberately withholds HRV from Health Connect. We use HR as a proxy. CRS works but with lower confidence. Samsung Sensor SDK in Phase 2 fixes this.

Budget Watches (Noise, boAt)

HR only, no HRV, unreliable sleep staging. Basic CRS — stress detection limited. Still useful for morning briefs.

Agent OS — Full System Architecture

graph TB subgraph Input["Data Input"] AW["Apple Watch
(HealthKit)"] GW["Android Watch
(Health Connect)"] end subgraph Phone["On-Phone (Offline)"] NM["Native Modules
Swift / Kotlin"] SQL["op-sqlite + SQLCipher
(AES-256)"] CRS["CRS Engine
(TypeScript)"] STRESS["Stress Detector"] end subgraph Cloud["Supabase Backend"] PG["Postgres + RLS"] CRON["pg_cron (15min)"] EF["Edge Functions"] end subgraph Agent["Agent Core"] PB["Prompt Builder
(25 fields)"] HOOKS["Hook Pipeline
(10 hooks)"] HAIKU["Claude Haiku 4.5
(tool_use)"] TOOLS["8 MVP Tools
(expandable to 50+)"] end subgraph Memory["Memory System"] T1["Tier 1: Structured KV"] T2["Tier 2: Summaries"] T3["Tier 3: Semantic + KG"] end subgraph Delivery["Delivery"] TG["Telegram"] WA["WhatsApp"] PUSH["Push Notifs"] INAPP["In-App"] end AW --> NM GW --> NM NM --> SQL SQL --> CRS SQL --> STRESS CRS --> PG STRESS --> PG CRON --> EF EF --> PB PB --> HOOKS HOOKS --> HAIKU HAIKU --> TOOLS TOOLS --> HAIKU HAIKU --> HOOKS HOOKS --> TG HOOKS --> WA HOOKS --> PUSH HOOKS --> INAPP HAIKU --> T1 T1 --> T2 T2 --> T3 T1 --> PB classDef blue fill:#e0e7ff,stroke:#6366f1,stroke-width:1px classDef green fill:#dcfce7,stroke:#22c55e,stroke-width:1px classDef amber fill:#fef3c7,stroke:#f59e0b,stroke-width:1px classDef pink fill:#fce7f3,stroke:#ec4899,stroke-width:1px class Input blue class Phone green class Agent amber class Memory pink
OpenFang: Hands + Prompt BuilderPaperclip: Heartbeat + PARAOpenViking: L0/L1/L2 tiersCoPaw: Hooks + SkillsPi Mono: Compaction

CRS — The Brain

Cognitive Readiness Score. Grounded in SAFTE-FAST (US Army biomathematical fatigue model, peer-reviewed). Hursh et al., Aviation Space Env Med, 2004

CRS = Sleep(0.35) + HRV(0.25) + Circadian(0.25) + Activity(0.15)

ComponentInputsKey Logic
Sleep (35%)Duration, stages, efficiency, debt7h target, 49h/week, ±2h bedtime deviation = -20
HRV (25%)RMSSD (Apple), HR proxy (Samsung)Personal baseline, time-of-day normalization (afternoon dip is normal)
Circadian (25%)Chronotype + time of dayPeak windows: early 6-9am, normal 10am-1pm, late 8-11pm. -40 outside peak.
Activity (15%)Steps, sedentary, exercise8K steps target, <4K = -30 penalty. Exercise context for stress masking.

All baselines are personal, not population norms. No commercial wearable CRS is peer-reviewed or algorithm-transparent. Bent et al., npj Digital Medicine, 2025

Stress Detection — Weighted Confidence

graph TD HRV["HRV Drop
weight: 0.35"] --> CONF["Weighted
Confidence"] HR["HR Elevation
weight: 0.25"] --> CONF DUR["Duration >10min
weight: 0.20"] --> CONF ACT["Inactivity Context
weight: 0.20"] --> CONF CONF --> C1{">= 0.60"} C1 -->|Yes| ALERT["ALERT"] C1 -->|No| C2{"0.40-0.59"} C2 -->|Yes| LOG["LOG ONLY"] C2 -->|No| IGN["IGNORE"] ALERT --> COOL{"2h cooldown?"} COOL -->|Clear| PRI["Risk Priority
Signal x Impact x
Timing x Novelty"] COOL -->|Active| SUPPRESS["SUPPRESS"] PRI -->|"> 0.40"| SEND["SEND"]

8-Point False Positive Prevention

1. Personal baselines — your normal, not population average

2. Time-of-day normalization — afternoon HRV dip is biological, not stress

3. Activity exclusion — exercising? HR spike is expected, not stress

4. 10-min persistence gate — single spikes are noise

5. 2h cooldown between alerts — max 3/day

6. Require cardiac signal — HRV or HR, not just inactivity

7. User feedback loop — thumbs up/down tunes thresholds (0.02 change rate)

8. Rules pre-filter — CRS > 60 AND conf < 0.3 → skip Claude entirely ($0)

The Agent Loop

sequenceDiagram participant C as pg_cron participant R as Rules Pre-Filter participant P as Prompt Builder (25 fields) participant H1 as Pre-Hooks (5) participant AI as Claude Haiku 4.5 participant T as Tools (8) participant H2 as Post-Hooks (5) participant TG as Telegram C->>R: check-triggers (every 15 min) R->>R: CRS > 60 AND conf < 0.3? alt SKIP (60-80% of checks = $0) R->>R: Log only else INVOKE R->>P: Assemble context P->>P: Soul + Zone + Mode + Tools + Memory + Biometrics P->>H1: Pre-reasoning hooks H1->>H1: Emergency / Gates / Inject / Compact / Rate H1->>AI: Send (2500 token budget) loop Max 3 iterations, 50s timeout AI->>T: tool_use call T->>AI: Result end AI->>H2: Response H2->>H2: Language / Confidence / LoopGuard / Memory / Analytics H2->>TG: Deliver end
OpenFang: Loop Guard (SHA256)CoPaw: Pre-Reasoning HooksHumanLayer: 40-60% Utilization RulePicoClaw: Provider Failover

25-Field Prompt Builder

Adapted from OpenFang's PromptContext (25 fields, 14 sections). Cache-optimized: stable system msg + dynamic user msg = near-100% cache hits on system prompt. Anthropic auto prompt caching, 2026

System Message (cached, stable)

#FieldTokens
1-4Soul base + zone modifier + mode template + safety rules~500
5Tool definitions (3-8 dynamic subset)~400
6-8User profile + core memory excerpt + health baselines~300

User Message (fresh per call)

#FieldTokens
9-11Current CRS + biometric snapshot + trigger context~300
12-13Last interaction + pending followups~200
14-15Conversation history (3 turns) + calendar~500

Total: ~2500 tokens max. U-shaped attention: critical rules at START, biometrics at END. Stable content in middle. From Agent-Skills-for-CE: "lost-in-middle" effect — middle positions drop to 76-82% recall accuracy

3-Tier Memory with Decay

graph TB subgraph T1["Tier 1: Structured (Always Loaded)"] ID["Identity"] HP["Health Profile"] PR["Preferences"] GO["Goals + Ancestry"] IN["Insights + Patterns"] end subgraph T2["Tier 2: Summaries (On-Demand)"] SS["Session Summaries"] PL["Pattern Log (append-only)"] PF["Pending Followups"] WC["Weekly Compaction"] end subgraph T3["Tier 3: Semantic (Phase 2)"] VEC["pgvector Embeddings"] KG["Knowledge Graph"] end T1 --> T2 --> T3 subgraph Decay["Memory Decay"] HOT["HOT 7d"] WARM["WARM 30d"] COLD["COLD 30d+"] end IN --> Decay

Key Patterns

Temporal Validity

Every memory entry: learned_on, last_validated, confidence, confidence_decay. Stale patterns auto-demote. From Paperclip PARA: hot/warm/cold decay

Record First, Answer Second

User says "started magnesium" → agent calls update_memory BEFORE responding. Context never lost. From CoPaw AGENTS.md

Close Every Loop

pending_followups tracks every suggestion. Morning brief checks outcomes. 3+ successful → promote to validated pattern. From Agency-Agents Sales Coach

L0/L1/L2 Health Summaries

L0 (~20 tok): "High-stress day, CRS 42". L1 (~100 tok): component breakdown. L2 (on-demand): raw data. 83% token reduction. From OpenViking benchmarks: LoCoMo10

Personality Spectrum + 4-Phase Nudge

5 Zones x 4 Modes = 20 Voice Combinations

ZoneCRSVoiceMessage Length
Energized80+Coach — challenges you3 lines
Steady60-79Friend — warm, informative3 lines
Flagging40-59Advisor — honest, protective2-3 lines
Depleted<40Caretaker — gentle, minimal1-2 lines
Crisis/No DataHonest — admits uncertaintyAdaptive

Zone selected by CRS + data quality. Mode selected by trigger type. From Agency-Agents: Whimsy Injector + Brand Guardian patterns

Every Proactive Message

Phase 1: Discover Preference

Track message style, timing, intervention type, cognitive load tolerance. First 14 days + ongoing.

Phase 2: Deconstruct to Micro-Action

"Take 3 slow breaths" not "improve your stress management". <2 min, zero willpower.

Phase 3: Deliver One Thing

[Signal] + [Meaning] + [Action] + [Buttons]. Max 3 lines. CRS <40 → max 2 lines.

Phase 4: Celebrate + Close Loop

Follow up 2h/24h later. Acknowledge effort not outcomes. Off-ramp always OK.

AI Onboarding Interview — Cold Start Killer

Not a form. The agent's first act of intelligence. Dynamic questions that adapt based on answers. Writes directly to core_memory. Same agent, different soul file (SOUL_ONBOARDING).

Progressive Profiling (Waves)

Day 0: Core Interview (3-5 min)

Goals, chronotype, medications, communication style, stress triggers. Dynamic branching — "better sleep" triggers sleep follow-ups, not generic questions.

Day 3: Sleep Deep-Dive (2 min)

"You slept 5.8h last night. Is that typical?" — uses real wearable data to make questions contextual.

Day 7: Stress Calibration (2 min)

"Your HRV dropped at 2pm. What was happening?" — triggered by first real stress detection.

Day 14: Relationship Check-In

"How's OneSync working? What should I do differently?" — course-correct with 2 weeks of data.

What Makes This Novel

Implicit Behavioral Calibration

Agent learns HOW you communicate during the interview itself. Terse answers → brief mode. Detailed → data-rich. Emotional → warm. Calibrated before first morning brief.

Medical Safety via Conversation

Naturally surfaces medications (beta-blockers affect HR), pregnancy (changes HRV), conditions. A checkbox gets skipped. A conversation gets honest answers.

Cold Start Solved

Agent has a useful profile from Day 1 — not Day 14. Goals, chronotype, meds, style all set before the first morning brief arrives.


WHOOPOuraNoriOneSync
OnboardingStatic formStatic formStatic formAI interview
Adapts to answersNoNoNoYes
Progressive wavesNoNoNoDay 0/3/7/14
Calibrates personalityNoNoNoFrom interview

Tech Stack

LayerTechnologyWhy This Choice
MobileReact Native + Expo SDK 53+Cross-platform from Day 1. Custom dev client for native modules.
iOS NativeSwift (HealthKit)HKHeartbeatSeriesQuery → beat-to-beat IBI → true RMSSD. Gold standard HRV.
Android NativeKotlin (Health Connect + WorkManager)15-min background sync. Samsung HR proxy for missing HRV.
Local DBop-sqlite + SQLCipherAES-256 on device. Health data encrypted at rest. Not WatermelonDB (perf), not Realm (lock-in).
StylingNativeWind v4Tailwind for RN. Not Gluestack-UI (too heavy), not StyleSheet (too verbose).
StateZustand + TanStack Query + MMKVClient + server + KV. No Redux complexity.
BackendSupabase (Postgres + Edge Functions + RLS)Row-level security built in. pg_cron + pgmq. Free tier → $25/mo at 200 users.
AIClaude Haiku 4.5 (@anthropic-ai/sdk)Messages API + tool_use. Direct calls, no middleware, no Agent SDK (stateless Edge Functions).
BotgrammY (Deno)Telegram bot. Webhook-based. Free API forever.
Schedulingpg_cron + pgmqBuilt into Supabase. Morning briefs, stress checks every 15 min.

Data Confidence Tiers

TierSourceAgent Language
HIGHApple Watch (beat-to-beat IBI → RMSSD)"Your HRV dropped 25%"
MODERATEPixel Watch, Fitbit (algorithm varies)"Your HRV appears lower"
LOWSamsung (NO HRV in Health Connect, HR proxy)"Based on heart rate patterns..."
DEGRADEDMissing, stale >6h, watch off"I'm missing data — how are you?"

Samsung HRV gap confirmed via Health Connect API testing, March 2026. Deliberate Samsung omission, not a bug.

Cost Per Claude Call

2000 cached tokens × $0.0001/tok$0.0002
1000 fresh tokens × $0.001/tok$0.001
350 output tokens × $0.005/tok$0.00175
Total per call~$0.003

Anthropic Haiku 4.5 pricing: $1/$5 per MTok (input/output). Cache reads: 0.1x.

MVP — What We Ship First

Phases A → H (Gate-Based)

PhaseDeliverableGate
AWoO (5-7 users, manual CRS)3+ say morning briefs useful
B1/B2HealthKit + Health ConnectCRS updates every 15min, real data
CDashboard (CRS gauge, sleep, trends)Real data, auto-updating
DAgent Core + Telegram (8 tools)Bot responds with health context
EProactive: morning brief + stress alertsBrief at wake time, alerts work
F-HOnboarding, 14-day self-test, beta<5min onboard, FP <20%, 5-7 users

8 MVP Tools

get_crs
get_sleep
get_stress_events
get_activity
get_user_profile
read_memory
update_memory
send_message

MVP Unit Economics

Free tier~$0.27/user/mo AI costAcquisition
Pro ₹399/mo ($4.34)~$0.90/user/mo79% margin
Break-even~50 Pro subscribers
Total MVP build cost$966 (vs $50K-$150K industry avg)

Proactive Intelligence — 4 Levels

Each level requires the previous to work. This is the roadmap from health bot to cognitive co-pilot to autonomous agent.

Level 1: Reactive (MVP)

"Your HRV dropped 22% in the last 30 min."

Responds to current state. Threshold-triggered.

Level 2: Pattern-Aware (Phase G+)

"It's Monday 1:30pm — your HRV typically drops now. Break before next meeting?"

Recognizes YOUR recurring patterns.

Level 3: Predictive (Post-MVP)

"Sleep debt + 4 meetings tomorrow → CRS will crash by 2pm. Reschedule hardest meeting to morning?"

Forecasts future state.

Level 4: Autonomous (Phase 3+)

Auto-blocks calendar for recovery. Auto-declines low-priority meetings when depleted. Auto-adjusts notification cadence.

Acts with pre-authorized permissions.

OpenFang Hands: autonomous scheduled tasksPaperclip: goal ancestry + delegationSwarms: MixtureOfAgents for specialist synthesis

Beyond MVP — Workspace Intelligence

The agent knows your body. Next: it knows your world. Every workspace API is free.

📅 Calendar (Google/Outlook)

Read schedule → "You have 4 back-to-back meetings, CRS is 52. Want me to add a 15-min break at 2pm?"
Write → Auto-block recovery time when CRS predicted to crash.

📧 Email (Gmail/Outlook)

Scan for urgent items → Include in morning brief: "3 emails need response before 11am."
Suggest: "Your CRS is 87 right now — tackle the hard reply first."

💬 Slack / Teams

DND when CRS < 40. Auto-status: "In deep work — back at 2pm."
Surface: "You were mentioned in #engineering 3 times during your focus block."

✅ Task Manager (Linear/Todoist)

Priority reranking based on CRS: "Your CRS is 87 — tackle the P0 bug now, save admin tasks for afternoon dip."

🎵 Music (Spotify)

CRS-adaptive playlists: focus music when peak, calm ambient when flagging, nothing when depleted.

📚 Notes (Notion/Obsidian)

Auto-log health patterns to your knowledge base. "Added: Monday afternoon crashes correlate with back-to-back meetings."

Every integration = a new tool in the agent's toolkit. The 8 MVP tools expand to 50+ with workspace connectors.

Beyond MVP — Multi-Agent Specialists

One generalist agent → specialized health agents coordinated by an orchestrator. From Swarms: MixtureOfAgents + HierarchicalSwarm. From OpenFang: Hands per domain.

graph LR O["Orchestrator
(routes by context)"] S["Sleep Specialist
(Haiku)"] ST["Stress Specialist
(Haiku)"] A["Activity Specialist
(Haiku)"] N["Nutrition Specialist
(Haiku)"] SYN["Synthesizer
(Sonnet)"] O --> S O --> ST O --> A O --> N S --> SYN ST --> SYN A --> SYN N --> SYN SYN --> MSG["Final Message"]

Sleep Specialist

Bedtime optimization. Sleep debt tracking. Chronotype-aware scheduling. Circadian rhythm coaching.

Stress Specialist

Real-time intervention. Breathing exercises. Meeting spacing. Recovery protocols. Pattern discovery.

Activity Specialist

Movement nudges. Exercise recovery. Sedentary alerts. Step goals adaptive to CRS.

Nutrition Specialist

Caffeine timing. Hydration reminders. Meal impact on HRV. Alcohol/sleep correlation tracking.

3 Haiku specialists ($0.002 each) + 1 Sonnet synthesizer ($0.003) = $0.009/brief. 88% cheaper than single Opus ($0.075). Swarms MixtureOfAgents benchmark

Cross-Domain Possibilities

The Agent OS is domain-agnostic. The CRS + proactive messaging + learning memory pattern applies to every persona that cares about cognitive performance.

🎓 Students

Study session optimization. Exam prep scheduling based on sleep debt. "Your CRS peaks at 10am — do problem sets now, reading later."

🏃 Athletes

Training load vs recovery balance. HRV-guided intensity. "HRV is 15% below baseline — scale back today's session."

💼 Enterprise Wellness

Team-level CRS dashboards (anonymized). Meeting culture insights. "Your team's average CRS drops 18% on Wednesdays — too many all-hands?"

🩹 Clinical Research

Longitudinal HRV tracking for studies. Medication impact monitoring. Open-source CRS as a research instrument.

🚀 Founders / Executives

Decision quality scoring. Board prep optimization. "Your CRS is 38 — delay the investor call to tomorrow morning."

🏠 Remote Workers

Work-from-home burnout detection. Break scheduling. Social isolation alerts (activity + HRV patterns).

Same Agent OS. Different soul files. Different tool sets. Different domains. The architecture scales horizontally across personas.

Tool Ecosystem — 8 Today, 50+ Tomorrow

The Agent OS is tool-agnostic. Each tool is a function the agent can call via tool_use. Adding a new integration = adding a new tool definition.

MVP (8 tools)

get_crs
get_sleep
get_stress
get_activity
get_profile
read_memory
update_memory
send_message

Phase 2: Workspace (12+ tools)

get_calendar
block_time
get_emails
set_slack_status
get_tasks
prioritize_tasks
get_weather
play_music
log_caffeine
log_meal
get_trends
export_data

Phase 3: Autonomous (20+ tools)

decline_meeting
reschedule
draft_reply
snooze_notifs
create_task
delegate_task
book_focus_time
order_food

Phase 4: Multi-Agent (10+ tools)

spawn_specialist
query_kg
predict_crs
population_compare
generate_report
train_model
federated_learn
export_research

Dynamic tool loading: only 3-8 loaded per call. From HumanLayer: "Vercel reduced 17 tools to 2 → 100% success rate (up from 80%), 3.5x faster." We load by intent, not by default.

What 14 Systems Unlock

We synthesized production-proven patterns from the best open-source agent projects + NVIDIA's enterprise agent framework.

OpenFang (Rust, 137K LOC)

Hands pattern: autonomous scheduled tasks. Triple-layer memory. Loop guard (SHA256). 25-field prompt builder. 16 security systems.

Paperclip (AI company OS)

Heartbeat execution. PARA memory with decay (hot/warm/cold). Goal ancestry. Adapter pattern. Wakeup coalescing.

OpenViking (ByteDance)

L0/L1/L2 context tiers: 83% token reduction. Filesystem-paradigm memory. Intent-driven retrieval.

CoPaw (Alibaba)

Pre-reasoning hooks. Proactive recording. Bootstrap onboarding. Skills system with progressive loading.

Pi Mono + Production Agent Platform (enterprise)

Layered architecture. Auto-compaction. Steering queues. Brain → Orchestrator → Execution. Learning flywheel. Workspace files as config.

Agency-Agents (100+ personas)

Personality Spectrum. Behavioral Nudge Engine. Quality gates. Healthcare compliance. Communication styles.

context-hub (Andrew Ng)

Progressive disclosure. Data confidence tiers. Annotation-style learning. BM25 relevance.

NemoClaw (NVIDIA)

Versioned blueprints with plan-apply-rollback. Declarative per-hand policies. Operator-in-the-loop escalation. Multi-model routing middleware. A/B testing infrastructure.

+ Agent-Skills-for-CE, PicoClaw, OpenClaw, Swarms, HumanLayer

Token budgets. U-shaped attention. Provider failover. Multi-channel routing. MixtureOfAgents. 40-60% utilization rule.

NemoClaw Patterns — Phase 2 Infrastructure

From NVIDIA NemoClaw — enterprise agent governance framework. Deferred to Phase 2 (premature for 5 MVP users), but critical at scale.

Versioned Blueprints (plan-apply-rollback)

All agent config (soul files, tools, thresholds, routing) stored in agent_blueprints table. Every change is versioned. Roll back instantly if a new soul file causes bad messages. A/B test different personalities. Full audit trail of what changed when.

Declarative Per-Hand Policies

Each Hand (stress monitor, morning brief, sleep coach) has explicit allowed tools, data scopes, and constraints enforced at runtime. Stress monitor can only read CRS + send alerts. It cannot update memory or access calendar. Principle of least privilege per autonomous task.

Operator-in-the-Loop Escalation

High-stakes recommendations (suggest doctor visit, alert emergency contact, reschedule important meeting) require explicit user approval. "Ask the first time, remember the preference." Builds trust while maintaining safety.

Multi-Model Routing Middleware

Route by complexity + cost: Haiku for routine (40%), Sonnet for nuanced (25%), DeepSeek for cost-sensitive (20%), Opus for complex synthesis (3%). Provider failover on errors. Cost tracking per model per user.

Config Runs (Plan → Apply → Observe)

Every config change follows: plan (preview what changes) → apply (deploy to subset) → observe (monitor metrics) → promote or rollback. Same pattern as Terraform for infrastructure. No surprise regressions.

Why Phase 2, Not MVP

34 hours of governance infrastructure for 5 users is premature. MVP hardcodes everything. Once we have 50+ users and need A/B testing, rollbacks, and multi-model routing — NemoClaw patterns become essential. The architecture is designed for this expansion.

Security Architecture

Health data is sensitive. These are non-negotiable.

MVP (7 systems)

1SQLCipher AES-256All health data encrypted on device
2Supabase RLSauth.uid() = user_id on every table
3Output validationBanned medical phrases scanned before delivery
4Emergency bypassSuicidal ideation → immediate safety response
5Audit loggingEvery agent invocation: model, tools, tokens, time
6Rate limitingPer-user message + cost caps
7No health data in logsconsole.log, Sentry, analytics — NEVER health values

Phase 2+ (5 additional)

8Taint trackingLabel data at source (HealthKit/HC), track through pipeline
9Loop guardSHA256 duplicate tool call detection
10Prompt injection scannerDetect override attempts in user messages
11Approval gatesHigh-stakes recs need explicit confirmation
12Session repairRecover from corrupted conversation state

Security patterns from OpenFang (16 systems). Health language rules from Agency-Agents healthcare compliance agent.

ALWAYS: "Not a medical device" disclaimer. NEVER diagnose. Wellness framing, not clinical.

The Full Vision — What OneSync Becomes

graph LR MVP["MVP
Health Bot + Telegram"]:::g --> P2["Phase 2
Cognitive Co-Pilot"]:::b --> P3["Phase 3
Autonomous Agent"]:::a --> P4["Phase 4
Personal OS"]:::p MVP --- D1["8 tools
1 channel
Reactive
Single agent"] P2 --- D2["20+ tools
Calendar + Email + Slack
Pattern-aware
Specialist agents"] P3 --- D3["50+ tools
Auto-schedule, auto-decline
Predictive CRS
Federated learning"] P4 --- D4["Agent network
Full life integration
Autonomous
Cross-domain personas"] classDef g fill:#dcfce7,stroke:#22c55e,stroke-width:2px classDef b fill:#e0e7ff,stroke:#6366f1,stroke-width:2px classDef a fill:#fef3c7,stroke:#f59e0b,stroke-width:2px classDef p fill:#f3e8ff,stroke:#8b5cf6,stroke-width:2px

The end state: an AI that knows your body, your work, your patterns, and your goals — and optimizes your entire day before you wake up. Not a health app. Not a productivity tool. A personal cognitive operating system that makes you perform at your best, every day, without you having to think about it.

What's Next — And What We Need From You

Immediate Next Steps

Step 1: Wizard of Oz Test (This Week)

5-7 people. Wear your watch consistently for 5-7 days. Shivansh manually computes CRS from your wearable data every morning and sends you a personalized brief on Telegram. You tell us: useful or not?

Step 2: Build Phase B (After WoO passes)

HealthKit connector (iOS) first, then Health Connect (Android). Real wearable data flowing into encrypted local DB. CRS updating every 15 minutes automatically.

Step 3: Agent + Beta (Phases D-H)

Claude-powered agent, Telegram bot, proactive messaging. 14-day self-test, then 5-7 user beta. Gate: morning brief open rate > 60%.

What We Need From You

Be a Beta Tester

Wear your watch to bed. Use it consistently for at least 7 days. Apple Watch or Pixel Watch / Fitbit gives us the best data. Samsung works too — just lower HRV confidence.

Give Honest Feedback

Tap the buttons. Tell us when a message is useless. Tell us when it's annoying. Tell us when it's actually helpful. The agent literally learns from your feedback — every thumbs-down makes it smarter.

Don't Sugarcoat It

If the morning brief feels like noise, say so. If the stress alert is wrong, say so. We'd rather kill a bad feature than ship one nobody uses. Build → Break → Fix → Repeat.

Cost to Build This

$966
Total MVP cost
(6-month runway)
50-150x
Cheaper than avg
health tech MVP
50
Pro users to
break even

Build Break Fix Repeat

Nothing is sacred. Everything evolves with the code.

5 Source-of-Truth Docs
Master Ref, Northstar, OnePager, Research, Agent OS
52+ Architecture Patterns
From 12 open-source agent projects
9 Claude Code Agents
Planner, QA-breaker, security, CRS validator...

Full docs: localhost:3333 • with AI Q&A chatbot built in

OneSync • The biological intelligence layer for the agentic economy