HealthLog/docs/self-hosting/document-sources.md

Document picker — Paperless-ngx and Papra

People who keep their paperwork in Paperless-ngx or Papra can search it from HealthLog and import the documents they pick: under "Link a document" on a visit, a vaccination or a condition, and from the Documents page. They search by document name, narrow by tag and date, tick what they want, and HealthLog fetches those files and stores its own encrypted copy, exactly as if they had been uploaded.

The picker is off until the operator names the instances HealthLog may contact.

Turning it on

List the origins in the server environment:

DOCUMENT_SOURCE_ORIGINS="https://paperless.example.com,http://papra.lan:1221"

The variable is on the docker-compose.yml whitelist, so setting it in .env is enough; restart the app container afterwards. Unset or empty, the settings card and every import button stay hidden and the routes answer 404.

Each entry is one scheme://host[:port], the grammar NOTIFICATION_PRIVATE_ORIGINS uses (see notifications):

• only http and https; scheme, host and port must match what a person enters (https://paperless.example.com:8443 and https://paperless.example.com are two different entries);
• no path, query, credentials, wildcard or CIDR; a person may still enter a base address with a path (https://example.com/paperless) as long as its origin is listed;
• the unspecified address, link-local and the cloud-metadata range can never be listed, and a listed name that resolves there is refused when HealthLog connects;
• a malformed entry is logged once at startup, reduced to scheme and host, and grants nothing; the valid entries beside it keep working.

Unlike the notification list, this list is the whole decision: an origin that is not listed is refused even when it is a public address. Removing an entry cuts off the connections that point at it on their next request; people can still delete their stored connection.

What each person sets up

Settings → Integrations → Document archives:

• Paperless-ngx: the base address and an API token (Paperless-ngx: My Profile → API token). Paperless-ngx 2.16 or later (API version 9).
• Papra: the base address, the organization id (from the Papra address bar, org_ followed by 24 letters and digits) and an API key with documents:read and tags:read. The connection test reads the tags too, so a key without tags:read is refused at save.

HealthLog tests the connection before saving it. A saved token is only ever sent to the address it was saved for: changing the address to another origin asks for the token again. The token is stored encrypted (AES-256-GCM, rotated with the other keys by scripts/rotate-encryption-key.ts) and is never shown again, not even in part.

What leaves the server, and when

Only requests the person starts: a search, the tag list for the filter, a connection test, and one download per picked document. There is no background sync and no polling. Every request goes to the listed origin only, with redirects refused, a 10-second limit for searches and 60 seconds for a file, and a size cap on every answer (1 MiB for a search page, the vault's per-file limit for a document).

Limits and rules

• Name search on Paperless-ngx uses its own title filter. On a Paperless-ngx that stores its data in SQLite, that filter ignores upper and lower case only for plain ASCII letters: searching röntgen does not find Röntgen. That is Paperless-ngx's behaviour, not HealthLog's; PostgreSQL-backed Paperless does not have it.

• A document deleted in HealthLog is recognised by its content during the 30-day undo window, so the same file from another archive is not stored again; after the purge only its ids are remembered.

• Imports count against the person's normal upload allowance (60 an hour) and storage quota.

• A document that is already in HealthLog is not stored twice. Each import is keyed by system, instance (the archive's origin) and document id, so the same id on two Paperless-ngx instances is two documents; a key imported before v1.39.3 carries no instance and matches any instance. These are the same rules the document token and scripts/import-documents.mjs follow.

• A document the person deleted in HealthLog is never brought back by the document token or the import script, however often they re-send it. The person can bring it back themselves: from Recently deleted on the Documents page for 30 days, or by picking it again in the picker, which says "In Recently deleted" or "Deleted in HealthLog" on such a row. Picking one still in Recently deleted restores that same document; picking one already removed for good imports it again.

• A picked document follows the person's own automatic AI reading setting.

• Only the owner of a record can connect an archive or import from it. People with shared access to a record, whatever their level, do not see the picker.

• Connections are not part of backups. After a restore on another server, each person reconnects (one form); the imported documents themselves are in the backup with the rest of the vault.

• Delete all data and account deletion remove the connection and its token.

• A stored connection stays visible under Settings, read-only, when this variable is unset or the person switched Documents off, so it can still be removed.