A local-first family health archive that combines Obsidian templates, AI prompt workflows, and a TypeScript preprocessing pipeline for Apple Health exports.

Your health data stays on your machine. AI assists with structuring and analysis — you control what leaves your device.

• Overview
• Features
• Architecture
• Workflows
• Project Structure
• Installation
• Configuration
• Development
• Testing
• Troubleshooting
• Contributing
• FAQ
• License

AI Health Vault is designed for families who want structured health records without surrendering privacy to commercial apps. Clone the repository, open the Obsidian vault, and use included prompts or Claude Code skills to transform photos, PDFs, and wearable exports into organized Markdown and CSV files.

Version 2 adds a production TypeScript runtime with strict typing, optional Redis-backed export job caching, structured logging, and a full build/lint/test pipeline.

flowchart TB
    subgraph Inputs
        A[Photos / PDFs]
        B[Apple Health Export]
        C[Manual Notes]
    end

    subgraph Processing
        D[AI Prompts & Skills]
        E[Apple Health CLI]
        F[(Redis Cache)]
    end

    subgraph Storage
        G[Obsidian Vault]
        H[CSV Tracking Files]
    end

    A --> D
    B --> E
    C --> G
    D --> G
    E --> H
    E -. optional .-> F
    G --> H

flowchart LR
    subgraph Application
        CLI[CLI Entry Point]
        SVC[Apple Health Service]
        CFG[Config Module]
        LOG[Structured Logger]
    end

    subgraph Persistence
        REDIS[Connection Manager]
        CACHE[Export Job Cache]
    end

    CLI --> CFG
    CLI --> SVC
    CLI --> REDIS
    SVC --> LOG
    REDIS --> CACHE
    SVC --> CACHE

flowchart TD
    A[Clone repository] --> B[Open vault/ in Obsidian]
    B --> C[Add family members to hub page]
    C --> D[Capture health document]
    D --> E[Send to AI with prompt or skill]
    E --> F[Paste structured output into vault]
    F --> G{More records?}
    G -->|Yes| D
    G -->|No| H[Review trends in tracking CSVs]

sequenceDiagram
    participant User
    participant CLI as Preprocessor CLI
    participant Parser as SAX Parser
    participant Redis as Redis Cache
    participant Vault as Obsidian Vault

    User->>CLI: export.zip + output directory
    CLI->>Redis: Check cached job (optional)
    alt Cache hit
        Redis-->>CLI: Completed job metadata
    else Cache miss
        CLI->>Parser: Stream XML records
        Parser-->>CLI: Typed CSV files
        CLI->>Redis: Store job result (optional)
    end
    User->>Vault: Attach CSV summaries to member archive

ai-health-vault/
├── src/
│   ├── cli/                  # Command-line entry points
│   ├── config/               # Environment configuration (Zod-validated)
│   ├── errors/               # Typed application errors
│   ├── logging/              # Structured JSON logger
│   ├── redis/                # Connection manager and export job cache
│   └── services/
│       └── apple-health/     # Streaming preprocessor implementation
├── tests/
│   ├── fixtures/             # Sample Apple Health XML
│   └── unit/                 # Vitest unit tests
├── vault/                    # Obsidian vault template
├── prompts/                  # Portable AI prompt library
├── .claude/skills/           # Claude Code skill definitions
├── guides/                   # Setup guides and FAQ
└── docs/internal/            # Engineering audit notes

Design decisions

• src/services/ holds domain logic; infrastructure lives in src/redis/, src/config/, and src/logging/.
• Content assets (vault/, prompts/, .claude/) remain separate from application code.
• Tests mirror the src/ layout under tests/unit/.

• Node.js 20 or later
• Obsidian (recommended for vault management)
• Redis 6+ (optional, for export job caching)

git clone https://github.com/runesleo/ai-health-vault.git
cd ai-health-vault
npm install
cp .env.example .env

Build and verify:

npm run validate

Open the vault:

1. Launch Obsidian → Open folder as vault
2. Select the vault/ directory
3. Edit 健康管理中心.md with your family member names

Copy .env.example to .env and adjust values:

# Development (tsx)
npm run preprocess -- --input /path/to/export.xml --output ./csv-output

# Production build
npm run build
npx apple-health-preprocess --input export.zip --output ./csv-output --types heart_rate,step_count

Supported output types: heart_rate, resting_heart_rate, walking_heart_rate_average, heart_rate_variability_sdnn, oxygen_saturation, vo2max, step_count, sleep_analysis, workouts.

# Type check
npm run typecheck

# Lint
npm run lint
npm run lint:fix

# Format
npm run format:fix

# Run tests in watch mode
npm run test:watch

# Full validation pipeline
npm run validate

1. Create a prompt file in prompts/
2. Add a matching skill in .claude/skills/
3. Update the vault hub page with workflow instructions
4. Document field mappings in vault/知识库/推荐字段标准.md

npm test

Test suites cover:

• Apple Health type parsing and CSV generation
• Output directory safety checks
• Redis export job cache read/write operations
• Redis connection manager configuration guards

Run the full pipeline before submitting changes:

npm run validate

The CLI refuses to write into a directory that already contains files. Provide a new path or clear the target directory first.

1. Confirm Redis is running: redis-cli ping should return PONG
2. Verify REDIS_URL matches your instance
3. Set REDIS_ENABLED=false to run without caching

rm -rf node_modules dist
npm install
npm run validate

Refer to vault/知识库/推荐字段标准.md and include it as context when prompting your AI assistant.

Use forward slashes or quoted paths:

npm run preprocess -- --input "D:/exports/export.xml" --output "./output"

1. Fork the repository and create a feature branch
2. Run npm run validate before committing
3. Keep commits focused and write descriptive messages
4. Update documentation when changing user-facing behavior
5. Open a pull request with a clear summary and test plan

See CODE_OF_CONDUCT.md and SECURITY.md for community and security guidelines.

Do I need Redis? No. Redis is optional and only caches export job metadata.

Which AI tools work? Any model that accepts text prompts: Claude, ChatGPT, Gemini, or local models via Ollama or LM Studio.

Is my data sent to the cloud? Only if you paste it into a cloud AI service. The vault itself is local Markdown and CSV.

Can I use this without Obsidian? Yes. Prompts, skills, and CSV outputs work with any editor.

Where is the detailed setup guide? See guides/快速开始.md and guides/FAQ.md.

MIT — Copyright (c) Leo (@runes_leo)