Portolan is a new foundation for sharing geospatial data: cloud-native files on object storage — no servers, no databases, no proprietary licenses. Structured metadata and open formats mean an agent can read a catalog, understand it, and analyze it directly. Open, AI-first, easy to implement, scalable, low cost, and sovereign.
SDIs are becoming more relevant than ever — but their requirements are changing.
100% open source under Apache-2.0, open formats, open governance. Geospatial data is data: DuckDB, BigQuery, and Pandas read the same bytes as desktop GIS. If Portolan disappeared tomorrow, your data would still work everywhere.
GeoParquet · COG · STACToday's SDIs weren't built for AI. Portolan explains to agents in plain text files how to access the data — no API calls needed — and structured metadata tells them what it means. The Finland.SDI demo shows an agent answering questions against a national SDI.
AI as a first-class citizenA traditional SDI needs databases, services, and specialist staff — a barrier most organizations can't clear. A Portolan node is files in a bucket: point, convert, push.
point · convert · pushAgents crawl entire datasets, not single queries. With no servers, scaling is handled entirely by cloud storage — the most robust, proven layer any cloud offers.
0 servers · ∞ requestsOnly two costs: storage and egress. Sharing public data shouldn't be an organizational burden, and popular datasets shouldn't cause budget shortfalls. The cost calculator shows what your data would run.
storage + egressControl the full stack within your own jurisdiction — no foreign vendor in the stack. Host on AWS, GCS, Azure, MinIO, Hetzner, Scaleway, or any S3-compatible storage.
bring your own bucketA community-built set of formats eliminates the need for a server — and scales through the power of cloud storage. No database, no tile server, no portal software: browsers, query engines, and AI agents read the files directly.
Transform shapefiles, GeoTIFFs, WFS, GeoPackage, and ArcGIS Feature Services into cloud-native formats: GeoParquet for vectors, COG for rasters, PMTiles for map tiles.
Generate a Portolan catalog that describes every asset with structured metadata. It is built on STAC, so any STAC tool can read it.
Push the files to any S3-compatible object storage: AWS, GCS, Azure, R2, MinIO, or Source Cooperative for free open data hosting.
Open the result in the Portolan browser or any STAC viewer, query it with DuckDB, or point an agent at the catalog URL — data answers questions directly.
Convert, validate, and sync. One command takes a folder of shapefiles to a browsable catalog on S3.
A lightweight viewer that reads Portolan catalogs and cloud-native files directly, with no server behind it.
Read moreSkills that let Claude, Gemini, and Codex agents publish catalogs and query cloud-native geospatial data.
Read moreBrowse an existing catalog, then publish your own data with the CLI or with an AI agent.
Portolan catalogs are live on the web. Open the browser to explore published datasets, or point DuckDB at any GeoParquet URL and query it in place.
Install portolan-cli, then init, add, check, and push. The CLI converts your files and generates the catalog.
Install the Portolan skill in Claude Code and point it at your data. It handles conversion, metadata, and publishing — AI tools excel at CLIs and text files, so non-experts can publish too.
Portolan is early-stage and developed in the open. Start here — the demos, the cost calculator, and the example catalogs are all linked from inside the talks.
“An SDI without servers.”
Five requirements for a modern SDI, the cloud-native formats that meet them, and a live serverless portal for the Netherlands.
“SDIs were built for experts; now they must serve agents.”
Why de-intermediation makes SDIs more important than ever, and the architecture they need — including the live Finland.SDI agent demo.
Discover and connect to Portolan catalogs from around the world.
Submit your Portolan catalog to the registry.