Health Connect connector

Slug health_connect. Health Connect is Android's on-device health store: apps such as Samsung Health, Google Fit, Fitbit, Oura, OMRON connect and CGM apps write to it, and an app with the user's permission reads it back. It has no server API, so an Android app reads it on the phone and uploads the records to POST /v1/ingest/health-connect. No keys are needed; the connector is always on.

Status: available on the server. The Kotlin SDK that reads Health Connect and uploads in this format is being built; until it ships, your Android app implements the contract below.

Upload contractCopy Anchor Link

The route works exactly like the Apple Health upload: a user token with the upload scope, the headers X-OHS-Type, X-OHS-Day, X-OHS-Batch and X-OHS-Part, and a gzip NDJSON body of at most 64 MiB. Re-sending a part is safe. Use the record type of the lines as X-OHS-Type, and keep (batch, part) unique across the types of one day.

gzip -c records.ndjson | curl -s -X POST $OHS/v1/ingest/health-connect \
  -H "Authorization: Bearer $USER_TOKEN" \
  -H 'X-OHS-Type: BloodGlucose' -H 'X-OHS-Day: 2026-03-02' \
  -H 'X-OHS-Batch: 7f3c' -H 'X-OHS-Part: 0' \
  --data-binary @-

Line formatCopy Anchor Link

One Health Connect record per line, as a JSON object. Field names follow the androidx androidx.health.connect.client.records classes, with the unit in the name.

• recordType: the record class name without Record (BloodGlucoseRecord is BloodGlucose). Required.
• metadata: id (required: the Health Connect record id), dataOrigin (the package name of the app that wrote it), lastModifiedTime, clientRecordId, recordingMethod (unknown, active, automatic or manual) and device (type, manufacturer, model).
• Instant records: time (an RFC 3339 instant, as Instant.toString() writes it) and an optional zoneOffset (+02:00).
• Interval records: startTime, endTime, startZoneOffset, endZoneOffset.
• Enum values are lowercase strings: the constant name without its prefix (SPECIMEN_SOURCE_INTERSTITIAL_FLUID is interstitial_fluid, STAGE_TYPE_DEEP is deep, EXERCISE_TYPE_BIKING is biking).

Example lines:

{"recordType":"BloodGlucose","metadata":{"id":"0b1e...","dataOrigin":"com.freestylelibre.app.de","recordingMethod":"automatic"},"time":"2026-03-02T07:30:00Z","zoneOffset":"+01:00","levelMillimolesPerLiter":5.6,"specimenSource":"interstitial_fluid","relationToMeal":"fasting"}
{"recordType":"BloodPressure","metadata":{"id":"9a2c...","dataOrigin":"jp.co.omron.healthcare.omron_connect","device":{"type":"unknown","manufacturer":"OMRON","model":"M7"}},"time":"2026-03-02T07:00:00Z","zoneOffset":"+01:00","systolicMillimetersOfMercury":121,"diastolicMillimetersOfMercury":79,"bodyPosition":"sitting_down"}
{"recordType":"Deletion","recordId":"0b1e..."}

A line of another record type is stored and skipped, so an app can send more than this list; unknown fields are ignored. A line that is not a JSON object, or has no recordType, refuses the whole part with 400. A supported record without metadata.id or with unreadable times fails at processing.

Mapping notesCopy Anchor Link

• Source: provider is health_connect; source_slug names the app that wrote the record (metadata.dataOrigin). Known apps get their usual slug (com.dexcom.* is dexcom, com.freestylelibre* is freestyle_libre, com.ouraring.* is oura, Mi Fitness, Mi Health and Zepp Life are xiaomi, Zepp (Amazfit) is zepp, RingConn is ringconn, ...); any other app gets its package name as a slug. Source priorities can rank them like any other source.
• Edits: Health Connect records can be edited in place. A sample is identified by its record id and start, not its value, so a re-sent record replaces the stored value, and a record whose start moved replaces its older copy. Heart rate series points are identified by time and value.
• Deletions: send the DeletionChange record ids from the changes API as Deletion lines. The stored records of that id are removed and never written again.
• Glucose: specimenSource goes to metadata.specimen. When it is interstitial_fluid (a CGM), the FHIR output codes the Observation with LOINC 99504-3 instead of 2339-0.
• Blood pressure: both halves share time, source and device, so the read API and FHIR pair them into one panel.
• Sleep: stages awake and awake_in_bed are awake, sleeping counts as light, light, deep and rem keep their name; out_of_bed and unknown give no phase.
• Menstruation: flow light, medium and heavy keep their name as the label, with the values Apple Health uses (2, 3, 4); unknown or no flow is unspecified (1). The samples feed the menstrual cycles. They are identified by record id and time: a re-sent record is a no-op, and an edited flow keeps the stored value until a Deletion of the record. MenstruationPeriod (a span with no flow per day) is not stored.
• Time zone: zoneOffset (or startZoneOffset) gives metadata.utc_offset_minutes and, for whole hours, metadata.timezone, which sets the local day of daily summaries.
• Manual entries: recordingMethod manual sets was_user_entered.

Reading permissionsCopy Anchor Link

An app reads only the record types the user granted. By default it reads back data from 30 days before its first permission grant; older data needs the history read permission (PERMISSION_READ_HEALTH_DATA_HISTORY).