Daslab Gateway
Some systems will never be reachable from the internet, and the ones worth asking about are usually the ones with the strictest answer. Daslab Gateway is one binary you run on a machine inside that network. It holds a read-only copy of the system, answers questions against the copy, and reaches Daslab over a single outbound connection. You open no inbound port.
The gateway is in pilot. A node is set up together with us; there is no self-serve install. The gateway page is the shorter version. This one is for whoever has to run it.
It reads SAP through the data preview, as an account you create
For an SAP system the node sends its SQL through the data preview in ABAP Development Tools, the endpoint developers use to look at table contents, as a named account you create for it. No database user, no database port, so there are no database credentials to leak.
That is a security answer, not a licensing one. SAP's API Policy does not list the data preview among the routes it approves for this kind of access, so check the node against your SAP agreement before you run it.
Every statement it sends is a SELECT. There is no code path in the binary that writes to the source, and the account you create for it needs no write authorisation, so the guarantee can hold on both sides.
What lands on the node
Two things, in separate shapes.
Rows land as Parquet files, one per batch. Each row carries the key it was fetched under and the time it was observed, and the newest observation of a key wins, so re-reading a page you already hold changes nothing. That is what makes a sync safe to interrupt and safe to repeat. From time to time the node folds a table's batches into one file holding the newest version of every row, deletions included, and removes the batches it folded.
Documents land as files next to the business object they belong to. The content of an attachment is not reachable from SQL in most systems, so it comes through an OData service your system already publishes for its own apps, and the catalogue that records it is an ordinary table you can query alongside the rest.
A query from a scene reads the copy. It does not reach your source system.
Checking the copy against the source
On a schedule, the node compares its copy against the live system, table by table, and within each table by date: a row count and a total on each side. Where the two disagree it re-reads just that slice and checks again, then records which tables match and when it last looked.
gateway_health returns that: the tables that differ, and how old the check is. Past about a day and a half it reads as unverified rather than healthy, because a stopped sync looks identical to a current one until something compares them.
What it costs the system it reads
Two ceilings on document fetching, both set by you and enforced in the binary rather than by convention:
| Setting | Default | What it caps |
|---|---|---|
DOC_RATE | 4 | requests per second, shared across every worker |
DOC_MAX | 500 | documents fetched per run |
The second one matters more than it looks. A node that can work through an unbounded backlog unattended is a node that can become a load event on a production system at four in the morning. With a per-run cap it makes steady progress instead, and you can raise it once you know what the source can take.
Every SQL read and every question is logged on the node
Reads against the source: the statement, the SAP user that sent it, whether it ran, the row count, and how long it took.
Questions from a scene: the statement, whether it was answered or refused, how many rows and bytes went back, and how long it took. A question is logged under the node's access token, not the person who asked; who asked is in the scene's history.
Both go into one append-only file on the node, a line per event, which a SIEM can ingest and which you can query with SQL. Refusals are logged as well as successes. Document fetches are recorded in the document catalogue, not in this file.
That is the log whoever administers the source system asks for first, and it is what makes a read-only claim checkable rather than something you take on trust.
Where it appears
Once connected, the node is an account in your workspace, with the same choice as any account of who can use it. A scene you add it to can query it, and the questions and answers land in that scene's history like any other work.
Updated 2026-11-14