Pulling documents

Orders, invoices, payments and returns are all pulled the same way: by version, never by page number. Every change to a record gives it a higher version, and versions are handed out in commit order, so a cursor can’t skip a record that was being written while you read.

GET /orders?filter[version_after]=0&per_page=100
→ {"data": [ …oldest change first… ],
   "meta": {"per_page": 100, "has_more": true, "next_version_after": 18342, "high_water": 18990}}

GET /orders?filter[version_after]=18342&per_page=100
→ … until "has_more": false

The loop, on a schedule

1. Read your watermark: the highest version you have staged, or 0 the first time.
2. GET /{orders|invoices|payments|returns}?filter[version_after]=<watermark>&per_page=100.
3. Stage every row under a unique key (id, version) — and line_number for line tables. A record that changes while you read comes back later with a higher version; the unique key makes the second copy harmless.
4. Repeat with meta.next_version_after until meta.has_more is false.
5. Book from staging, then ack what you booked.

Tip

Book one sales order per Slflo order, and before creating it look for an existing one carrying the order’s id (D365: CustomersOrderReference). If your job died after creating the sales order but before the ack, the next run finds it instead of creating a twin.

Filters

Every collection takes these:

Parameter	Meaning
filter[version_after]	The cursor.
filter[updated_after]=2026-10-01T00:00:00Z	First load only: start at the first change from that time. Not with a cursor.
filter[ack]=pending	Only records not acked at their current version. Also accepted, rejected, any (default).
filter[id]=a,b,c	Only these ids — at most 100 (more answers 400 INVALID_REQUEST).
filter[external_ref]=SO-0012345	Only the record you acked under that key.
per_page	1–500, default 100. page= and offset= are refused.
filter[status]=ERP_PENDING,ERP_FAILED	Orders only.
filter[seq_after]	Invoices, payments and returns only — see the invoice counter.

Brackets may be sent raw (filter[ack]) or encoded (filter%5Back%5D). An unknown filter[…] is refused, not ignored, so a typo can’t silently widen a query.

Recovering after a lost cursor

If your staging table is purged or the ERP database restored, the watermark is gone. Don’t start from 0 and re-book: pull filter[ack]=pending instead. It returns exactly the records you haven’t acked at their current version — including ones that changed after you acked them.

What each document carries

• Order: header; customer, branch and rep with both keys; lines (product, unit, quantity, price, discount, free-goods flag, promotion) with VAT per line worked out as the invoice will; totals; status; idempotency_key; and your last ack.
• Invoice: as issued — lines, VAT, totals — its order with that order’s lines, and its state with the tax authority (tax_submission).
• Payment: method (CASH, CHEQUE, TRANSFER, INSTAPAY, WALLET), amount, allocations to invoices, the receipt, the cheque and its state, and a reversal when undone.
• Return: lines against the original invoice’s lines, the reason, the approval, and the credit note number.

Try it

The loop in curl and jq, printing each order’s id, version and number. -g stops curl reading the brackets as a pattern:

v=0
while :; do
  page=$(curl -sg "$SLFLO_BASE/orders?filter[version_after]=$v&per_page=100" -H "Authorization: Bearer $SLFLO_TOKEN")
  echo "$page" | jq -r '.data[] | "\(.id) \(.version) \(.number)"'
  [ "$(echo "$page" | jq .meta.has_more)" = true ] || break
  v=$(echo "$page" | jq .meta.next_version_after)
done
echo "watermark: $v"

The same in Python, with staging and booking, is the pull connector.