Hackathon project brief
Apple Health context. Private by design.
Vital Agent Sync gives authorized AI agents fresh, scoped Apple Health context through an iPhone app, a user-controlled local receiver, and standard MCP tools.
Model scaling determines what AI can do.
Context scaling determines whether it can do it for me.
01 / The problem
AI can understand the world, but it does not know today’s you.
“Based on last night’s sleep and my current state, am I ready for a hard workout?”
Without current personal evidence, an Agent does not know how long the user slept, whether the data is fresh, or whether its assumptions match the user’s real state. It can only produce a generic answer.
02 / The product
A user-controlled path from Apple Health to an authorized Agent.
The Agent never accesses HealthKit directly. The iPhone syncs authorized summaries, and the Agent asks for the minimum necessary context through MCP.
- 01
Apple Health
The iPhone reads only the HealthKit categories the user authorizes.
- 02
iPhone summaries
The app prepares compact summaries instead of giving an Agent direct HealthKit access.
- 03
Local receiver
Encrypted sync lands in a user-controlled SQLite database on the user’s machine.
- 04
MCP
Standard tools return scoped context with source and freshness metadata.
- 05
Authorized Agent
The Agent receives the evidence needed for a question, not blanket health-data access.
03 / Working proof
The Agent gets evidence, not HealthKit permission.
The working loop runs across a physical iPhone, an encrypted local receiver, SQLite, and MCP: configure the runtime, scan the local pairing QR, approve Health permissions, sync, and ask an Agent a personal-state question.
- Every relevant result identifies when the source data was last updated.
- Observed Apple Health facts are kept separate from limited Agent inference.
- Missing or stale data is reported instead of silently filled in.
- The connection can be inspected and revoked by the user.
04 / Why this exists
A more personal AI should not require surrendering all personal data.
Apple Health already contains a rich record of daily life, and modern models are capable of using that evidence. What is missing is a safe, continuous, controllable connection between the two. Existing paths often mean repeated manual exports or moving the entire dataset into a central platform. Vital Agent Sync tests a different bargain: useful context without transferring ownership.
05 / Core thesis
The next scaling axis is context.
Model scaling
- Reasoning
- Multimodality
- Tool use
- Capability
Context scaling
- Identity
- Memory
- Real-world state
- Freshness
- Permission
Context scaling is not a longer context window. It is the ability to continuously acquire, select, verify, and safely use real-world context.
06 / Design principles
Context should scale without transferring ownership.
- 01
Users choose which personal data sources to connect.
- 02
Each authorized agent receives only the context needed for the current task.
- 03
Returned context includes source and freshness information.
- 04
Users can inspect, revoke, and replace connections.
07 / Current status
Local Preview is working; the product boundary is explicit.
Implemented now
- A physical iPhone can pair with the receiver and sync user-authorized Apple Health summaries.
- The local runtime stores summaries in SQLite and exposes them through MCP tools.
- WorkBuddy users can install the Skill from SkillHub and start the guided LAN setup from a conversation.
- The npm package provides the same runtime for WorkBuddy, Hermes, OpenClaw, and generic MCP clients.
- Trusted LAN is the default route; a user-managed Tailscale network is the optional private remote route.
Not claimed yet
- The iOS app is a development build and is not yet distributed through the App Store.
- The Local Preview relies on a foreground iPhone sync; it does not promise a fixed background schedule.
- Hosted relay is experimental and is not required, available, or recommended for the current onboarding path.
- Vital Agent Sync is not a medical service and does not diagnose, prescribe, or replace professional care.
08 / Direction
Health is the starting point, not the boundary.
The long-term direction is a user-owned context layer that can connect more personal data sources to more authorized Agents while preserving scope, provenance, freshness, auditability, and revocation. Vital Agent Sync begins with Apple Health because sensitive health context makes those requirements concrete from day one.
09 / Clear answers
Frequently asked questions
- Does an AI Agent receive HealthKit permission?
- No. HealthKit access stays in the iPhone app. An authorized Agent can only call scoped MCP tools against summaries already synced to the user-controlled receiver.
- Does local storage mean the model is local?
- No. The database and receiver stay on the user’s machine, but context returned through MCP may be sent to the model provider selected in the Agent. Users should choose that provider and scope deliberately.
- What does context scaling mean here?
- It does not mean a longer prompt. It means continuously acquiring, filtering, verifying, and safely using relevant real-world context while preserving provenance, freshness, permission, and user control.
- Why begin with Apple Health?
- Health data makes freshness, sensitivity, provenance, and revocation impossible to ignore. It is a demanding first source for testing a broader user-owned context layer.