Hackathon submission update: HealthGraph Agent: Apple Health to Aura
quick update on my submission. Since my last post I've added a full set of Neo4j Aura elements on top of the original ETL + longevity-query foundation. The whole stack now runs end-to-end: iPhone → Aura graph → Aura Agent → iOS chat. Everything is open source and reproducible.
→ Repo + README: GitHub - ma3u/healthgraph-agent: Apple Health data → Neo4j knowledge graph with longevity-focused analysis. Neo4j Aura Agent Hackathon 2026. · GitHub
The submission is now built on four pillars, each backed by a Neo4j Aura primitive:
1. 
Aura Agent (HealthGraph Agent)
A longevity-focused assistant with 6 tools — Text2Cypher + 5 parameterized Cypher templates (health_overview, workout_recovery, longevity_trends, overtraining_check, exercise_balance). MCP enabled, REST-invokable.
The big addition: the agent is defined as code via the new Aura v2beta1 /agents API (thanks to Ed Sandoval for the pointer). The full agent JSON is committed at agents/healthgraph-coach.json and reconciled with one script (scripts/create_aura_agent.py) supporting three modes: status (default — diff live vs file), --pull (live → file), --push (file → live, POST or PUT). I keep a screenshot of the Console state in the repo for fidelity.
While wiring the iPhone chat I caught and fixed a copy-paste bug in the health_overview Cypher template — it referenced $workout_type while the tool parameters were $start_date/$end_date, so the agent was reporting "tool is currently unavailable". The rewritten template now returns daily metrics (RHR / HRV / steps / sleep_hours / workout_min / kcal / VO2max) for the requested window plus a rolling 30-day baseline ending the day before — exactly what the system prompt asks for.
Three real Q&As are captured at docs/AGENT_DEMO.md (in the repo): a grounded RHR analysis ("56.11 bpm all-time vs 57.82 bpm last-30"), a 12-week overtraining check with CAUTION flags + deload-week recommendations, and a "no data" failure case where the agent honestly reports a gap instead of fabricating.
2. 
Aura Dashboard
Whoop-style NeoDash dashboard at neodash/whoop_dashboard.json — 5 pages, 35 panels: daily hero card (Recovery %, Strain 0–21, Sleep %), Recovery deep-dive, Strain deep-dive, Sleep deep-dive, and 8.5-year Health Monitor. Score formulas in docs/SCORING.md. Pushed into Aura's built-in Tools → Dashboards via scripts/upload_dashboard.py — idempotent (deterministic UUID per title).
3. 
Aura GraphQL Data API + GitHub Pages
GraphQL Data API: curated SDL with @cypher MERGE mutations (ingestDay / ingestWorkout / ingestSleep) at cypher/graphql_schema.graphql. Deployed against the live tenant via scripts/create_aura_data_api.py — aura-cli v1.8.0 has no data-api commands, so the script hits the v1beta5 REST endpoints directly. All three mutations smoke-tested + idempotent.
GitHub Pages: daily Recovery snapshot rendered by scripts/render_snapshot.py, committed by a GitHub Actions workflow on a 06:30 UTC cron (auto-resumes paused Aura instances). The live link is in the header above.
→ Live daily snapshot: HealthGraph dashboard
4. 
iPhone App: HealthKit sync + "Ask your graph"
HealthGraphSync iOS app — Swift 6 / iOS 26.5 SDK. Reads HealthKit on-device, queries Aura for max(Day.date), scans HealthKit since then, presents the per-type delta, then uploads via the three GraphQL @cypher mutations. Includes Rescan last 30 days and Rescan last 365 days flows. Verified end-to-end on real iPhone 17 Pro — graph grew from 3,087 → 3,117 :Day nodes after first sync.
The Dashboard tab embeds an "Ask your graph" panel that calls the same HealthGraph Agent /invoke endpoint via OAuth2 client-credentials. Four suggestion chips compute concrete date ranges at tap time (ISO Mon–Sun for "last week", last 12 weeks for overtraining, last 30 / 60 days for workout-HRV correlations) so the agent always gets specific dates. The fresh answer auto-opens in a draggable overlay (.medium / .large detents) on top of the dashboard, with full Markdown rendering via AttributedString(markdown:) — paragraphs, bullets, numbered lists, inline **bold** / *italic* / `code`, and the trend arrows (↑ improving, ↓ declining, → stable) the agent emits.
BYO Aura
The whole thing is bring-your-own-Aura: every installer points the iOS app and the GitHub workflow at their own Aura instance. There's no shared backend. Both provisioning scripts (create_aura_data_api.py, create_aura_agent.py) take AURA_CLIENT_ID / AURA_CLIENT_SECRET / AURA_INSTANCEID from a local .env and produce a working stack in <5 min.
What's open
-
#5 Auth0 production sign-in path (Apple / Google / GitHub / Microsoft → Bearer JWT to the Data API). Currently using dev-mode
x-api-key; Auth0 work is gated on a manual tenant setup. -
Vector embeddings for similarity search on
DailySummary.description.
Issues closed during this stretch
#2 Apple Health Sync to Aura + In-app Dashboard · #3 GraphQL Data API + Pages dashboard · #4 Aura Agent integration + Neo4j Skills · #6 HealthGraphCoach as code via Aura v2beta1 /agents.
Feedback welcome 
— particularly on whether the agent's tool selection feels right, and on the v2beta1 agents-as-code workflow (super useful, please keep going with it).
Matthias / @ma3u