GeoParquet to PMTiles Converter

Convert one or two uploaded polygon GeoParquet layers into one downloadable PMTiles v3 archive containing MVT vector tiles. Use it for small static vector-map layers and Apify workflows without installing a local tiling toolchain.

You get MAP.pmtiles and an OUTPUT JSON report with tile counts, selected attributes, source IDs missing at each requested zoom, and the archive's SHA-256 hash. Each completed run also writes the report as one dataset row.

This Actor accepts WGS84 polygons only, with a combined limit of 16 MiB and 2,000 features. Tiling simplifies geometry and small polygons can disappear. It provides a downloadable archive; styling and permanent map hosting are outside its scope.

Upload and convert

1. Use the official Apify Python client to create a key-value store and upload an uncompressed binary record, as shown below. Set content_encoding="identity": the client's default compressed upload is unsupported. Keep the store ID and record key, such as areas.parquet.
2. Turn Use bundled lake example OFF. It is on by default and uses a Natural Earth lake instead of your files.
3. Select the uploaded-file store, confirm rights to process nonpersonal data, and describe each layer. Choose zooms suitable for your layer's extent.
4. Start the run. Download the archive after the run is SUCCEEDED and OUTPUT.status is success. Check the downloaded file's SHA-256 against OUTPUT.sha256 before using it.

Upload with the official apify-client 3.2.1 Python package and your own Apify API token in the APIFY_TOKEN environment variable:

import os
from pathlib import Path
from apify_client import ApifyClient

client = ApifyClient(token=os.environ["APIFY_TOKEN"])
store = client.key_value_stores().get_or_create(name="my-map-inputs")
client.key_value_store(store.id).set_record(
    "areas.parquet",
    Path("areas.parquet").read_bytes(),
    content_type="application/octet-stream",
    content_encoding="identity",
)
print(store.id)

Use that printed ID as inputStoreId. This API route is tested; a Console upload route is not verified. See Apify's create-store API and upload-record API. HTTP-compressed records are refused before conversion. Parquet's internal column compression is supported and is a separate input limit.

For a file with a unique area_id column and a scalar name column:

{
  "useDemo": false,
  "inputStoreId": "YOUR_STORE_ID",
  "rightsConfirmed": true,
  "layers": [
    {
      "name": "areas",
      "recordKey": "areas.parquet",
      "idColumn": "area_id",
      "attributes": ["name"]
    }
  ],
  "minZoom": 0,
  "maxZoom": 4
}

For small local building footprints, zooms 12 to 14 can be more suitable than the generic 0 to 4 example. Choose zooms for your own extent and check the omission report.

For two layers, add a second object to layers with a distinct name. Both files must be in the selected store. Output features include source_id as a string, plus your selected attributes. IDs must be unique within each input layer; output fragments can repeat the same source ID across tiles.

Open the downloaded archive locally in a PMTiles-compatible viewer, such as pmtiles.io, to check its appearance. In MapLibre, set source-layer to the output layer name shown in vector_layers[].id. Vector tile coordinates use Web Mercator (EPSG:3857); input and archive bounds use longitude/latitude. source_id is an attribute, not a unique MVT feature ID.

Inputs and limits

• GeoParquet 1.0 or 1.1 with binary WKB geometry, a declared Polygon or MultiPolygon geometry type, planar edges and WGS84 longitude/latitude coordinates. Projected or undefined CRS, points, lines and 3D geometry are unsupported.
• One or two layers, 16 MiB combined file size, 2,000 total features and 750,000 coordinate vertices. These are input limits, not a promise that every admitted request completes.
• A unique, nonempty string or integer ID per feature. Up to eight selected scalar attributes per layer: strings, booleans and finite numbers, with null values allowed. Dates and nested values are unsupported. Selected integers must fit the exact JSON integer range; use the ID column for larger integer identifiers.
• Zooms 0 to 14 and at most 5,000 potential tiles. The estimate uses each layer's bounding box with a buffer, so wide extents can be refused even for small files.
• At most 64 MiB declared uncompressed Parquet data, 16 MiB final archive, 256 MiB scratch space checked at intervals and ten minutes processing. Verification can also refuse unusually large fields or tile fragments.

An unsupported or over-budget request returns OUTPUT.status: refused with a reason and no archive. Storage or lifecycle failures can fail the run without a completion report. A report alone does not confirm completion: require the successful run and matching archive hash.

What the report checks

The converter reads the PMTiles directory and each addressed output tile using PMTiles and MVT libraries. It checks tile addresses, decoding, polygon fragment validity and winding, source IDs and selected attribute values. The report lists retained and missing source IDs separately for every requested zoom.

These checks describe the generated tiles. They do not certify geographic accuracy, seam appearance or lossless geometry. Tippecanoe clips, simplifies and quantizes polygons; coordinates near the poles are clamped to Web Mercator. Use your own map preview before relying on the appearance.

Price

US$0.25 per completed uploaded-file archive, whether it contains one or two layers, plus Apify's standard US$0.00005 start charge per GB of run memory. At the default 2 GiB memory setting, a successful uploaded-file conversion costs US$0.25010 in total. Apify usage for the run is included. Your own key-value store with the uploaded files follows your Apify plan. Set your run's maximum total charge to at least US$0.251.

The bundled example and refused requests have no conversion fee; the start charge still applies. The conversion charge is attempted only after the archive, report and dataset row are stored. A late billing or lifecycle failure can leave a FAILED run with a charged archive retained. Contact support before starting again in that case. Resurrection of the same run checks its billing record and retained archive without a second conversion charge. Resurrection can add another Apify start charge. A new run is a new job and can incur a new charge.

Files and privacy

Use only geographic data you have permission to process and that contains no personal records. This Actor reads your uploaded records and writes new outputs without changing the source files. It does not fetch arbitrary source URLs. Your files and outputs are stored on Apify under your storage permissions and retention settings. Deleting the run or its storage can remove access to outputs.

FAQ

Can I use it with a large national dataset? Only if your supplied layer fits all limits. This version serves small polygon jobs; it is not a bulk-data service.

Does it include a map viewer or hosting? No. Download the archive for a PMTiles-compatible map workflow and configure its styling and hosting separately.

Can I do this locally? Yes. Free local conversion tools are available. This Actor packages a bounded conversion job and a tile report for Apify workflows.

Where can I get help? Use this Actor's Issues tab or support@madrasco.dev. Platform replies may be AI-assisted; the human owner is available on request.

Built and maintained by Madrasco with AI agents. AI agents wrote the code and this page; scripts perform the conversion checks. Excludes safety-critical use, including navigation and survey decisions.

Changelog

• 0.1: initial release: one or two uploaded polygon layers, PMTiles vector output and per-zoom source-ID reporting.