Fast, lightweight, self-hosted geocoding engine in pure Rust.

630 million addresses in a single file.

Forward, reverse and batch geocoding built from OpenStreetMap and OpenAddresses, served from your own hardware.

Self-hosted
og runs on your own hardware. No address leaves your network, and there is no per‑lookup fee.
Fast
0.72 ms a query and 14,087 queries a second on the Netherlands: 6× Photon and Pelias.
Lightweight
One binary, one file, no database. The whole Netherlands fits in 337 MB.

Benchmarks

Paid-API accuracy. None of the weight.

On real web addresses, og puts more answers in the right city than TomTom or OpenCage, and trails TomTom by 1.4 points at street level. Against open-source engines it isn't close: the whole Netherlands on 1 vCPU and 1 GiB, at 0.72 ms a query and 6× the throughput of Photon and Pelias.

Read the full benchmarks →

Geocoding real web addresses

First answer within 20 km, % of queries (higher is better)

Geocoder	GH4	GH6
Google Maps	98.6%	88.3%
open-geocode	93.9%	78.0%
TomTom	91.8%	79.4%
OpenCage	90.2%	68.2%
Pelias	85.9%	67.8%
Photon	84.9%	62.8%
Nominatim	66.6%	55.6%

MESSY STREETS · 959 web addresses · other rows from the paper · og on its Planet Packmethod

Serving a country

Requests per second with p99 under 100 ms (higher is better)

Geocoder	Forward	Reverse
open-geocode	14,087	65,606+
Pelias	2,323	2,406
Photon	2,194	3,098

Every OpenStreetMap address in the Netherlands · each engine alone, memory capped · 16 vCPU / 64 GiB · reverse capped by the load generatormethod

Answering one query

One request at a time · milliseconds (lower is better)

Geocoder	p50	p99
open-geocode	0.72	4.2
Photon	2.0	19.4
Pelias	4.7	9.6

Every OpenStreetMap address in the Netherlands · each engine alone, memory capped · 16 vCPU / 64 GiBmethod

Serving 100 requests a second

Least memory to serve 100 req/s, in GiB (lower is better)

Geocoder	GiB	vCPU
open-geocode	1	1
Photon	4	2
Pelias	8	1

Every OpenStreetMap address in the Netherlands · each engine alone, memory capped · page cache includedmethod

Quickstart

v0.1.0 on crates.io ↗

From an extract to answers in one binary.

$ cargo install open-geocode --features tui

og — ~/packs — 80×2401 / 05

Compare

No API bills, no database, no cluster to run.

Commercial APIs are quick to start with, but bill every lookup and see every address.

Open-source geocoders keep the data in-house, but need a database or search cluster. og keeps it in-house and runs from one file.

Feature	og	Commercial	Open source
Self-hostedAddresses stay on your network	Yes	No	Yes
No usage feesNo quotas or rate limits	Yes	No	Yes
InfrastructureWhat you run and maintain	YesOne binary, one file	YesNone, it's hosted	NoDatabase + search cluster
Batch geocodingCSV in, coordinates out	YesFree, unlimited	YesFree tier, then per address	NoNo batch endpoint
Open sourceMIT licensed, open data	Yes	No	Yes

Open source: Nominatim, Pelias and Photon. Commercial: Google, HERE, Mapbox and others.

FAQ

Questions, answered.

Common questions about hosting OpenGeocode, what it costs, and how accurate it is.

Read the documentation

How do I use OpenGeocode?

Build a Pack with og build, then og serve it. See the quickstart.

What hardware do I need to host it?

Very little. This page's search serves the whole planet, a 26.3 GB Pack, from a 4 vCPU VM with 7.6 GB of RAM, using about 0.5 GB of memory.

How much does OpenGeocode cost?

Nothing. No API key, quota or per-lookup fee; you only pay for your machine.

Does it work offline?

Yes. It only needs a connection to download data when building a Pack.

Can I geocode a whole CSV?

Yes, with open-geocode geocode batch. See batch geocoding.

How accurate is it?

It leads the open-source geocoders on the MESSY STREETS benchmark. See the benchmarks.

What about data and privacy?

It runs on your hardware, so no address leaves your network.

Is OpenGeocode open source?

Yes, MIT-licensed, on GitHub.

OpenGeocode: free, open-source, self-hosted geocoding