MedCue

English · 简体中文

MedCue is a native medication-management app for iPhone and Apple Watch. It helps people organize medication plans, record dose events, review adherence trends, monitor supply, and prepare concise information for follow-up visits.

MedCue supports medication safety and routine management. It does not diagnose conditions, prescribe medication, or replace a clinician or pharmacist.

• A focused Today timeline for recording taken, delayed, skipped, corrected, and reopened doses.
• Medication profiles with treatment periods, dose changes, multiple reminder times, package photos, and inventory estimates.
• Records, calendar review, adherence trends, risk review, and visit-summary export.
• Apple Watch, complications, notifications, and Live Activities backed by the iPhone as the primary source of truth.
• Optional HealthKit signals presented as contextual trends without diagnostic interpretation.
• Local OCR and barcode-assisted import with mandatory user review before persistence.
• Cloud and on-device medical-assistant modes with explicit consent, scoped context sharing, transport validation, and response safety boundaries.
• Simplified Chinese source localization is present in the iPhone app's String Catalog; complete localization across companion bundles remains tracked in Issue #20, Issue #21, and Issue #22.
• VoiceOver labels and stable accessibility identifiers on selected navigation and editor controls.

The repository separates domain logic from Apple-platform integration:

• swift-core/ contains pure Swift domain models, scheduling, adherence, trend, inventory, label, risk, and AI-safety logic.
• ios-app/MedicationAdherenceApp/ contains the SwiftUI application, SwiftData persistence, app commands, system-service adapters, Watch app, Watch widget, and Live Activity extension.
• Packages/LlamaFramework/ provides the local model runtime package boundary.
• tools/ contains reproducible validation and release-safety checks.

Application writes use explicit command and transaction boundaries. SwiftData commits complete before user-visible success or post-commit side effects such as notifications, Watch snapshots, and Live Activity updates. The former Live Activity URL dose-action handler is disabled; the legacy intent no longer writes a dose. GitHub Issue #2 tracks the remaining authorization regression gate before this boundary is release-ready.

• Medication data is stored locally with versioned SwiftData schemas and migration coverage.
• iPhone notifications, alarms, Lock Screen Live Activities, and Dynamic Island views use generic reminder text without medication names or doses. Notification dose actions require device unlock; Live Activities open the app for confirmation. See ADR-0003 and the remaining Issue #14 validation.
• Cloud AI remains opt-in and receives only user-authorized context scopes.
• Credentials are not stored in source control and release artifacts are checked for sensitive configuration.
• iOS remote endpoints are HTTPS validated, client-side redirects are rejected, and sessions use ephemeral storage.
• Local model files are integrity checked before use.
• AI output is finalized through medical-response boundaries before display or persistence.
• The assistant cannot create, stop, replace, or change medication or dosage.

• macOS with Xcode 26.5 or newer compatible with the project format
• iOS 17.0 or later
• watchOS target support for Watch features
• A personal or organization Apple Developer Team for physical-device signing

Open:

ios-app/MedicationAdherenceApp/MedicationAdherenceApp.xcodeproj

The source repository intentionally omits the local llama binary. Ordinary builds use the local-inference stub by default:

xcodebuild \
  -project ios-app/MedicationAdherenceApp/MedicationAdherenceApp.xcodeproj \
  -scheme MedicationAdherenceApp \
  -configuration Release \
  -destination 'generic/platform=iOS Simulator' \
  CODE_SIGNING_ALLOWED=NO \
  -jobs 1 \
  build

To link the real on-device model runtime, first provide a local llama.xcframework described in ios-app/MedicationAdherenceApp/Frameworks/README.md, then build with MEDCUE_ENABLE_LOCAL_LLAMA=1. The stub path verifies the rest of the source and integration boundary; it does not verify the real llama binary or on-device inference. Repository automation does not yet verify the binary's source or digest, so those checks must be completed separately before treating the framework as trusted.

For a physical device, select the same development team for the main app and its extensions, then run the shared MedicationAdherenceApp scheme from Xcode.

Run the portable domain suite:

cd swift-core
swift test

Run the complete native validation gate from a clean source clone without the local binary:

tools/verify-native.sh

The native gate covers domain tests, hosted persistence and application tests, primary-navigation and first-launch UI smoke tests, unsigned Release builds, Watch builds, project preflight checks, and sensitive-artifact assertions.

With a locally supplied XCFramework installed, set MEDCUE_ENABLE_LOCAL_LLAMA=1 to link it during the gate. Real inference additionally requires the ignored GGUF model and the explicit smoke procedure; neither the CI stub nor a successful link build proves real-model behavior.

.
├── .github/
├── cloudfunctions/
├── docs/
├── ios-app/
│   └── MedicationAdherenceApp/
├── Packages/
│   └── LlamaFramework/
├── swift-core/
├── tools/
├── AGENTS.md
├── CONTEXT.md
└── README.md

Development uses an Issue -> branch -> Pull Request -> CI workflow. Start with AGENTS.md, CONTEXT.md, docs/PROJECT_STATUS.md, and docs/DEVELOPMENT_WORKFLOW.md. docs/README.md indexes current architecture, test, operational, historical, and GitHub localization material. The Chinese entry point is README.zh-CN.md, and the bilingual GitHub workflow is recorded in docs/GITHUB_LOCALIZATION.md. Historical append-only logs are retained for audit but are not the active backlog.

MedCue is a medication organization and education tool. Its reminders, risk summaries, trends, imported text, and AI-generated content may be incomplete or incorrect. Users should verify medication decisions with qualified healthcare professionals and follow the prescription, label, and clinical guidance applicable to them.