NODES 2026Relate by Neo4j Hackathon

MyHealthCoach

My local health graph.
Not the cloud's.

Paper lab reports and Apple Watch, in one private Relate graph on my iPhone and my Mac. Claude answers through Relate's MCP server, read-only.

Use ← → or the dots below. Everything on this page is clickable.

The problem

Health data on paper, in silos

  • Lab reports on paper, wearables in their own apps
  • Klarbefund scans a report on the iPhone
  • Every value coded with LOINC, with the page and line it came from
Tap the phone

Relate on the iPhone

A graph database inside the app

  • Neo4j's embedded graph database, in Klarbefund
  • One health.rl file under iOS Data Protection (complete): unreadable while the phone is locked
  • Excluded from backups, never sent anywhere
Tap the phone

Export to my Mac

A copy on hardware I own

  • Paired by QR code: pinned certificate and secret token
  • Over Wi-Fi with Apple's TLS; a USB cable works with the command-line receiver
  • Into Klarbefund for Mac, sealed with AES-256 while it is closed
Tap the phone

Klarbefund for Mac · TestFlight

No scripts. Three steps.

  • Pair: scan the Mac's QR code with the iPhone, tap Send everything now
  • See: one Relate file in the app's sandbox, sealed with AES-256-GCM while the app is closed; Open 3D view shows it as an interactive graph, your main risk in the middle
  • Ask: Connect Claude Desktop installs a read-only extension with one click; the Klarbefund skill takes your personal anamnesis and stores the findings in your graph

TLS, hashing and signatures come from macOS itself: no bundled crypto. Same TestFlight link as the iPhone app.

Klarbefund for Mac with fictional data: pair your iPhone, your health graph, ask Claude

Metadata model

One graph, grown from one root

Knowledge graph

Health risk → score → measure → my value

The graph also holds the medical knowledge, each link with its source. Pick a risk:

Risk scores that estimate it

Measures the guidelines name

Ask Claude

Answers from my own graph

What is in my health graph? Count the kinds of data, the sources and the time span.
What risk scores are stored for me? Show the lab values each was calculated from.
Before my doctor's visit: write a six-minute briefing with my trends and the guideline targets.

Claude Desktop writes the Cypher; Relate's MCP server answers read-only from a snapshot.

From graph to action

Ready for my doctor's visit

  • Claude reads the graph through Relate's MCP server
  • Charts the trends, computes SCORE2, compares with the guideline targets
  • Writes a six-minute briefing with one question per topic, and its sources

The briefing, as Claude laid it out

1 · Statin: which LDL target applies to me?2.5 min
2 · Hair loss: which tests, which referral?1.5 min
3 · Vaccinations: what is due today?1 min
4 · Screening at 50: what to order?1 min

Values stay out of this public page; the video shows the real briefing.

Security

Private by construction

iPhone

Data Protection

Relate's single file under FileProtectionType.complete, excluded from backup. Talks only to the paired Mac.

Mac

AES-256 seal

Klarbefund for Mac keeps the graph sealed (AES-256-GCM) while it is closed; the key never leaves the Mac's keychain. Pinned TLS or a USB cable, no cloud on the way.

Claude

read-only

The Klarbefund extension reads a snapshot copy, only while the Mac app is open; write queries are refused. Only the rows Claude asks for leave the Mac.

For research

The data stays. The analysis comes to it.

Primary use · this app

My graph, not the cloud's

One person, their own data, on their own iPhone and Mac. Claude reads it read-only, for their own doctor's visit.

Secondary use · research

A Trusted Research Environment

Many people's graphs, for a researcher. The rows never travel to them: the work happens inside a sealed workspace. Under the EHDS, that's the secure processing environment (Art. 73).

How a TRE earns trust: the Five Safes

Safe projects

A permit for one purpose

Safe people

Named in the permit, second factor

Safe settings

Confidential container, attested, no internet

Safe data

Pseudonymised, opt-outs removed

Safe outputs

Only reviewed aggregates leave

For research

Klarbefund's graph in a researcher's remote desktop

The person Klarbefund on iPhone and Mac shares Access body permit: purpose, people, datasets sealed Attested workspace · no internet Researcher's remote desktop (Linux) ● Relate: one .rl per pseudonymised person ● health_graph: knowledge, scores, anamnesis ● the 3D view, served inside ● relate serve --mcp --read-only same code as on the Mac airlock Researcher reviewed aggregates only

Checked on Linux

Relate 1.0.1 for Linux in a container: all 67 tests pass, and the anamnesis plan, the 3D view and the read-only MCP server run, on the synthetic sample graph.

Planned, not built

A workspace per permit (#1132) and an airlock reviewed by the access body (#1133). Claude only through a model endpoint the access body approves.

Try it

Clone, run, see the graph

git clone https://github.com/ma3u/relate-hackathon-team
cd relate-hackathon-team
relate query --database examples/health-graph/health.rl --mode table <<< \
  "MATCH (h:HealthRisk)<-[:MARKER_OF]-(:LoincCode)<-[:CODED_BY]-(o:Observation)
   RETURN h.name AS risk, count(o) AS values"
relate serve --mcp stdio --database examples/health-graph/health.rl --read-only

The sample store is synthetic: no real person. README · Devpost · 3-minute version

QR code: Klarbefund on TestFlight

Klarbefund on TestFlight

iPhone and Mac

testflight.apple.com/join/ssADSXX6