libpostal: address normalisation in C, and the geocoder it refuses to be
A C library for parsing/normalizing street addresses around the world. Powered by statistical NLP and open geo data.
At a glance
- What is it?
- libpostal turns messy free-form street addresses into normalised strings using statistical NLP trained on open geo data, ships official bindings for six languages, and has not cut a release tag since v1.1 in 2018 even though commits landed in 2026.
- Who is it for?
- Adopt libpostal when your problem is deduplicating, indexing or comparing addresses that arrived as typed by humans, and when you already have a geocoder downstream. Do not adopt it expecting coordinates, and note that the README's opening states a goal of understanding location strings in every language rather than a measurement.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Activity is slowing. The repository last received commits 6 months ago.
- What is it written in?
- Mainly C, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- openvenues/libpostal4,926 ★ · +41
| Repository | 2026-09-11 | 2026-11-15 |
|---|---|---|
| openvenues/libpostal | 4.9k | 4.9k |
41 Stars in the last 65 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
The newest tag is from 2018 and the last push is from 2026
The release history and the commit history disagree by eight years. The most recent tags are v1.1, named Walla Walla, on 2018-05-09; v1.0.0, Brooklyn 99, on 2017-04-07; and v0.3.4, Quito, on 2017-02-09. The last push to master was on 2026-05-13, and the repository is not archived.
So the code moves and the version numbers do not. Whatever sits on master after v1.1 carries no tag that tells you which behaviour you are running, and the naming convention for releases, a place name attached to each one, has not been used since 2018.
That matters more here than in most projects. This is a statistical model: the parser's output depends on training data, and a change to data/ or versions/ can change addresses it parses correctly without any interface change to notice it. .bumpversion.cfg sits at the root, so version bumping is configured, yet no newer version was cut. Pin a commit hash, and record the data directory you installed alongside it.
It normalises addresses, and says plainly it is not a geocoder
The README states the boundary in one sentence: though libpostal is not itself a full geocoder, it can be used as a preprocessing step for a geocoding application. That sentence is the fastest way to decide whether you want this library.
What goes in is a free-form address, the kind a person typed. What comes out is a clean normalised form, described as suitable for machine comparison and full-text indexing. The value is in making two spellings of the same place comparable, so a search index or a duplicate detection routine stops treating them as different strings.
What you do not get is a coordinate or a verified location. The README's own list of what addresses matter for is place search, transportation, on-demand and delivery services, check-ins and reviews, and for every one of those the normalised string is an input, not an answer.
The problem is abbreviations, and full-text search cannot see past them
The reasoning behind the library is a complaint about search. Even the simplest addresses are packed with local conventions, abbreviations and context, which makes them hard to index and query with traditional full-text search engines. One street writes its type out, another abbreviates it, a third uses a local word, and the same building number appears in three shapes.
A full-text index treats those as unrelated tokens. Matching on them requires knowing which abbreviations mean which words in which country, which is exactly the knowledge a statistical model can absorb from data.
So libpostal is best understood as a normalisation stage for an index rather than as a search engine. If your problem is ranking documents that mention an address, feeding normalised forms into the index is the win. If your problem is resolving an address to a point on the map, this library is one step upstream of the answer and no closer than that.
The model is trained on open geo data, and the training set is in the tree
The method is statistical NLP over open data, and the research behind it is written up in two posts rather than inside the repository. The original is Statistical NLP on OpenStreetMap, and the follow-up covers the 1.0 release.
The tree shows how much of the project is data rather than code. There is a data/ directory, a versions/ directory, and a file called current_parser_training_set at the root. resources/, scripts/, src/, test/ and m4/ sit alongside them.
The consequence is that installing libpostal is more than compiling C. The model files are part of the deliverable, and an upgrade that changes them changes results. Anyone integrating this into a pipeline should treat the data files with the same care as a database migration, because that is what they are.
Six official bindings live in six other repositories
The core library is pure C, and the officially supported bindings are separate projects: pypostal for Python, ruby_postal for Ruby, gopostal for Go, jpostal for Java, php-postal for PHP and node-postal for NodeJS. The README says it is easy to write bindings in other languages.
That structure has a practical cost. There is no single install that gives you the library and your language's wrapper, and the C library has to be present before any of them works, which means a compiler and the data files on every machine that runs your service, including the ones your CI builds on.
The upside is that the interfaces are not frozen by the core library's own release cycle. A binding repository can move independently of the C library, which matters when the C library's newest tag is from 2018 while its commits are from 2026.
Autotools at the root, and Windows kept in its own directory
The build system is GNU Autotools, and the root shows every part of it:
configure.ac
Makefile.am
bootstrap.sh
libpostal.pc.inThe last of those is the template that produces a pkg-config file so other software can find the installed library.
Windows has its own arrangement. There is a windows/ directory, a libpostal.def for the DLL export list, and .appveyor.yml for continuous integration on that platform, alongside the GitHub Actions badge in the README.
What that layout means in practice is that a Unix build expects autotools to be present before anything else happens, and that a Windows build is a separate path rather than a flag.
Around the world is a goal, and no coverage table exists
The stated goal is to understand location-based strings in every language, everywhere, and the README runs through a long line of language flags to make that point. It is a goal statement, not a measurement.
There is no table of supported languages, no accuracy figure, no benchmark and no example of an input with its output anywhere in the opening text.
If your users span many countries with unfamiliar scripts and address conventions, test your own formats against it.
New language coverage is funded through OpenCollective
The project asks companies using libpostal to sponsor it, and explains why in terms of the research rather than the bug tracker: interpreting what humans mean when they refer to locations is far from a solved problem, and sponsorships help pursue new frontiers in geospatial NLP. Sponsors get their logo on the repository page with a link to their site, and the money runs through an OpenCollective page.
CODE_OF_CONDUCT.md, CONTRIBUTING.md and an ISSUE_TEMPLATE.md sit in the root, so bug reports have a channel.
Editorial conclusion
Adopt libpostal when your problem is deduplicating, indexing or comparing addresses that arrived as typed by humans, and when you already have a geocoder downstream. Do not adopt it expecting coordinates, and note that the README's opening states a goal of understanding location strings in every language rather than a measurement. Two things to check before you build on it: which commit you actually get, since the newest tag is v1.1 from 2018-05-09 while the last push was 2026-05-13, and whether the language mix of your addresses is covered by whatever training data ships in data/ and versions/.
Frequently asked questions
What is libpostal?
A C library for parsing and normalising street addresses around the world, using statistical NLP and open geo data. It converts free-form addresses into clean normalised forms suitable for machine comparison and full-text indexing, and the README states it is not itself a full geocoder.
how to use libpostal
As a preprocessing step in front of a geocoding application: free-form addresses go in, normalised forms come out, and those forms are what you index and compare. The README is explicit that libpostal does not resolve addresses to locations, so it leaves the geocoding step to you.
how to use libpostal in python
Through the officially supported Python binding, openvenues/pypostal, which wraps the pure C core. Bindings for Ruby, Go, Java, PHP and NodeJS live in their own repositories too, and the C library plus its data files have to be installed first.
how to install libpostal
The repository builds with GNU Autotools: configure.ac, Makefile.am, bootstrap.sh, an m4/ directory and libpostal.pc.in at the root, with a windows/ directory and libpostal.def for the Windows path. Bindings are installed from their own repositories rather than from this one.
is libpostal free
It is MIT licensed, with the licence file at the repository root. The project also runs an OpenCollective sponsorship page and asks companies that use it to sponsor, with a logo and link on the repository page in return.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/openvenues-libpostal)Star history chart for your README
[](https://hysenlabs.com/projects/openvenues-libpostal)Was this useful?