Folder catalog
File-based integration is the oldest there is and still the most common on the shop floor: a system drops a CSV on a network share and someone has to read it. The work is not in reading the file — it is in getting encoding, delimiter and header right, which nobody gets right first time and which only fail on the first real file.
This assistant answers those questions by looking at the file.
It opens from the same table icon as the database catalog, in External Connections or on an Application row — when the connection is of type File, what opens is the folder catalog.
Browsing
The catalog shows folders and files starting from the connection’s base directory, with a name filter. It is real navigation: you go into subfolders until you find the file that represents the case.
Unlike the database catalog, here the CMS reads content — the first few KB of a file — because there is no way to discover format, encoding and delimiter without looking. The screen says so explicitly, and nothing in the folder is changed.
What the detection finds
Once a file is chosen, the CMS analyses the sample and answers:
| Field | How it is decided |
|---|---|
| Encoding | BOM first (it is proof); then strict UTF-8 — Latin-1 text with accents produces an invalid sequence and gives itself away; alternating null bytes mean UTF-16LE; isolated nulls mean binary |
| Format | JSON, JSON lines, CSV, XML, text or binary |
| Delimiter | Tries ;, ,, tab and | and picks the one that separates consistently |
| Header | Whether the first CSV line looks like column names |
| Columns | Names from the header, or positional (campo1, campo2…) when there is none |
| Emission mode | Whole file in one message, or one line per message |
| Preview | The first lines already decoded, so you can check with your own eyes |
Detection is deterministic, without AI. The file answers for itself; a model’s guess here could only get it wrong — the same criterion as the database catalog.
Everything detected comes pre-filled and stays editable. If the sample misled it, you correct it before generating.
What you decide
| Choice | Default | Why |
|---|---|---|
| File mask | The extensions present in the folder | * would pick up the control file the other side uses to signal end of write |
| Emission mode | The detected one | One line per message gives per-record traceability; whole file preserves the document |
| Ignore header | The detected one | Only meaningful in per-line mode |
| Post-read action | MOVE | See below |
| Schedule | Cron | How often to sweep the folder |
The post-read action is born as MOVE, and is not guessed
After reading, the CMS can move, rename, delete or leave the file. The assistant always suggests MOVE.
MOVE takes the file out of the way, which gives idempotency from the file system itself: a file that was read is not read again, and it still exists if someone needs to check it.
DELETE is irreversible. Suggesting it from what was seen in a folder would be irresponsible: the assistant does not know whether that file is copied anywhere else. If your plant’s policy is to delete, tick it yourself — but let it be a decision, not a default.
The envelope, and the Transformer that comes with it
In per-line mode, the CMS delivers an envelope with the raw line:
{
"arquivo": "PRODUCAO_20260826.csv",
"caminho": "/dados/entrada/PRODUCAO_20260826.csv",
"linha": 42,
"conteudo": "000060003285;0010;480;12;ST"
}What splits the columns is a Transformer — and the assistant writes it, from the detected columns, with the right delimiter and header. That is why the catalog’s columns do not become the Collection’s contract: they become the material that Transformer is written from.
In whole-file mode, the message is the file content, with metadata attached.
Read only
The folder catalog generates only the Collection. Writing files (FILE_WRITE) depends on a
content model — which columns, in what order, with what separator, with or without a header — and
the folder does not reveal that. See Files to configure writing by hand.
Review and install
As in every assistant: the review shows each object with CREATE, UPDATE or CUSTOMISED, the codes
are editable, and the Collection is born disabled.