Exporting layers to files

Any cached layer can be downloaded as a file. Exports run as background jobs, so a large base converts without tying up the API, and one request can produce several formats at once.

The endpoint is shaped like Esri’s — POST .../FeatureServer/export, returning an esriJob… status document — so ArcGIS clients that already speak createReplica need no special handling.

1. From the UI

Open a base and pick the Export tab. Choose one or more formats, tick the layers you want, optionally set a projection or a filter, and hit Export. A progress card appears and turns into a row of download buttons when the job lands.

Downloads expire (24 hours by default, see Expiry), so treat the links as short-lived rather than permanent URLs.

2. Over REST

curl -X POST 'https://{SERVER}/api/v0/rest/services/{baseId}/FeatureServer/export' \
  -H 'Authorization: Bearer {key}' \
  -H 'Content-Type: application/json' \
  -d '{ "formats": "gpkg,shapefile" }'

{
  "jobId": "6f1c…",
  "jobStatus": "esriJobSubmitted",
  "statusUrl": "https://{SERVER}/api/v0/rest/services/{baseId}/FeatureServer/export/jobs/6f1c…"
}

Authenticate exactly as you would a FeatureServer read: a session cookie, an org- or base-scoped API key, or an app token. Parameters may be sent as JSON body fields or as query-string parameters — ArcGIS clients use both, so either works.

Scope

Path	Exports
`/rest/services/{baseId}/FeatureServer/export`	every table in the base
`/rest/services/{baseId}/FeatureServer/export?layers=0,2`	just those layer ids
`/rest/services/{baseId}/{tableName}/FeatureServer/export`	that one table

POST …/createReplica is an alias of …/export, for clients that only know the Esri name.

Poll for the result

curl 'https://{SERVER}/api/v0/rest/services/{baseId}/FeatureServer/export/jobs/{jobId}' \
  -H 'Authorization: Bearer {key}'

{
  "jobId": "6f1c…",
  "jobStatus": "esriJobSucceeded",
  "progress": 100,
  "expires": "2026-09-10T14:02:11.000Z",
  "results": {
    "paths": [
      { "format": "gpkg", "sizeBytes": 1841152, "url": "…/jobs/6f1c…/results/gpkg" },
      { "format": "shapefile", "sizeBytes": 903221, "url": "…/jobs/6f1c…/results/shapefile" }
    ]
  }
}

jobStatus is one of esriJobSubmitted, esriJobExecuting, esriJobSucceeded or esriJobFailed. Each paths entry is a direct download — same auth as everything else.

A job that produced some of the formats it was asked for still succeeds, and reports the rest under messages:

"messages": [{ "type": "esriJobMessageTypeError", "description": "dxf: ogr2ogr exited 1 …" }]

So always read messages even on esriJobSucceeded — never assume every requested format is in results.

3. Formats

Request them as a comma-separated formats list (or repeat format=).

Common

`format`	File	Notes
`geojson`	`.geojson`	Copied straight from the cache when unfiltered, so it is essentially free
`csv`	`.csv`	Geometry written as WKT
`xlsx`	`.xlsx`	One sheet per layer
`shapefile`	`.zip`	Field names truncated to 10 characters by the format itself
`gpkg`	`.gpkg`	One table per layer
`kml` / `kmz`	`.kml` / `.kmz`	One folder per layer

Esri

`format`	File	Notes
`filegdb`	`.zip`	A `.gdb` folder, one feature class per layer; opens in ArcGIS Pro
`featurecollection`	`.json`	Esri FeatureSet JSON with a `layerDefinition`
`spatialite`	`.sqlite`	SpatiaLite database — Esri’s “SQLite geodatabase”; QGIS reads it directly

Interchange

`format`	File	Notes
`flatgeobuf`	`.fgb`	Cloud-optimized; seekable and streamable
`geoparquet`	`.parquet`	Columnar, for DuckDB / Spark / pandas analytics
`gml`	`.gml`	OGC XML, common in European open data
`dxf`	`.dxf`	CAD exchange (AutoCAD); attributes are largely dropped by the format

Vector tiles

mbtiles builds a vector tileset rather than converting features, and takes its own options (zoom range, per-layer field keep-lists, projection) plus an optional snapshot block that publishes it at a hosted tiles endpoint. It does not accept the filters below. See the Vector tiles tab in the UI, and Vector tiles and snapshots for the full option set.

One file or many?

gpkg, filegdb, xlsx, kml, kmz, gml and spatialite hold many layers natively, so a multi-layer export is a single file with a layer/table/sheet per table.

The rest hold one layer each, so a multi-layer export of geojson, csv, shapefile, featurecollection, flatgeobuf, geoparquet or dxf arrives as a .zip of per-layer files. A single-layer export is always the plain file.

4. Filters

All four are optional, all apply to every layer in the export, and all are evaluated by GDAL:

Parameter	Example	Effect
`outSR`	`3857`	Reproject from 4326 to this EPSG code
`where`	`population > 100000`	SQL attribute filter
`bbox`	`-10,40,5,55`	Spatial filter, `minX,minY,maxX,maxY` in 4326
`outFields`	`name,population,iso3`	Keep only these attributes (`*` or omitted keeps all)

curl -X POST '…/{baseId}/FeatureServer/export' \
  -H 'Authorization: Bearer {key}' -H 'Content-Type: application/json' \
  -d '{
        "formats": "geojson,gpkg",
        "layers": "0,1",
        "outSR": 3857,
        "where": "population > 100000",
        "outFields": "name,population"
      }'

Because the filter is shared, a where clause or a field name that only exists on some of the selected layers will fail those layers — they show up in messages while the rest of the export still succeeds.

bbox is always interpreted in 4326, even when outSR asks for something else: it filters the source data, then the survivors are reprojected.

5. Expiry

Artifacts live on disk under the export id and are swept by an hourly maintenance job. The window is EXPORT_TTL_HOURS (default 24); the exact deadline for a given job is in its status document as expires. After that both the download and the status endpoint return 404, and a new job has to be submitted.

6. Exports read the cache

An export converts whatever Sync Server last synced, not the upstream source. A layer with no cache yet fails with No cache for layer "…" — sync the base first., so trigger a sync (POST /api/v0/bases/{baseId}/sync) and wait for it before exporting a newly added table.