GridPin

Offline forward and reverse geocoding from a memory-mapped country file. Off the grid. On the grid.

No server, no database import, no per-request fees, and every geocoding query stays on your machine. After you install the package and download a sheet, GridPin makes no network requests.

pip install --upgrade gridpin==0.5.0

The wheel contains the engine, not the data: download a country sheet separately. Free static sheets for the United States, Canada, Germany, France, Italy, Spain, Mexico, the Netherlands, Brazil and Serbia are listed at https://gridpin.dev/docs#delivery — no account, no key, no sign-up.

One command does the download and the check: pip install gridpin-mcp, then gridpin-fetch fr puts the France sheet in ~/.cache/gridpin and verifies it against the signed release; from Python, gridpin_mcp.sheets.download("fr") returns the same path. The same package is an MCP server for AI agents (uvx gridpin-mcp).

With pandas: pip install "gridpin[pandas]", then gridpin.sheets.fetch("us"), gridpin.geocode_df(df, "address", "us") (or df.gridpin.geocode(...)) and gridpin.reverse_df(df, "us"); the gp_* columns are the same as in gridpin batch CSV and gp_geocode_df in R. gridpin.sheets.list / path / verify / fetch use the same cache and checks as gridpin sheets. Extras [pandas], [geopy]; without them the wheel still has no dependencies.

The United States is one nationwide sheet: point addresses in 35 states, Census address ranges elsewhere; Canada is one nationwide sheet (all provinces and territories). Germany covers 15 of 16 Länder; Bavaria is not covered. Spain covers all 52 postal provinces; Brazil is one nationwide sheet; Mexico is one nationwide sheet. Client update required: use 0.7.0, because older engines have no Mexican rules.

import gridpin

g = gridpin.Geocoder("france.bin")
g.geocode("8 Boulevard du Port, Amiens", 1)
g.reverse(49.8945, 2.2949, 1)

# optional POI layer (cascade built in: address index first,
# POI only on weak results, exact addresses never overridden)
g = gridpin.Geocoder("france.bin", poi="fr_poi.bin")

If GridPin is useful to you, a star on GitHub helps other people find it.

Three limits worth knowing before you build on this:

• Reverse returns the nearest indexed address — or a street-centroid approximation — together with distance_m. It is not a rooftop fix.

• Speed: exact lookups run at ~3,000/s, measured on the France sheet and a single arm64 laptop core. Typo-tolerant fuzzy matching is much slower (tens of ms/query) — use geocode_many() to batch.

• POI is France-only today and opt-in. Address lookup is the core product; the POI layer is younger and never overrides an exact address match.

• Country files & docs: https://gridpin.dev/docs

• Quality benchmark, including where other geocoders score higher: https://gridpin.dev/bench

• Source: https://github.com/gridpin/gridpin

• Issues: https://github.com/gridpin/gridpin/issues

• Engine license: Apache-2.0. Data files carry their own source licenses — see https://github.com/gridpin/gridpin/blob/main/ATTRIBUTIONS.md