Transforms a GTFS file into a directed acyclic graph of actual connections.
A connection is the combination of a departure and its successive arrival of the same trip. Our goal is to retrieve a list of connections that is sorted by departure time, better known as a Directed Acyclic Graph. This way, route planning algorithms can be performed.
More information and live demo at https://linkedconnections.org
Install it using the Node Package Manager (npm). Node.js 24 or 26 is required.
npm install -g gtfs2lcIf you haven’t yet picked a GTFS file you want to work with, different repositories exist. Our favorite ones:
Yet, you may also directly ask your local public transport authority for a copy.
Mind that we have not tested our code with all GTFS files yet, and there are known limitations.
You can use your favorite unzipper. E.g., unzip gtfs.zip should work fine.
This process runs automatically, so you can skip to Step 4. You can also run it independently with gtfs2lc-clean <path>. The cross-platform cleaner normalizes the CSV files and uses a disk-backed external sort for stop_times.txt.
stop_times.txt is ordered by trip_id and stop_sequence; source files are only replaced after their normalized versions have been written successfully.
Successfully finished the previous steps? Then you can now generate actual departure and arrival pairs (connections) as follows:
gtfs2lc /path/to/extracted/gtfs -f jsonThe default output is a binary Jelly-RDF stream (linkedConnections.jelly). We support other formats such as jsonld, turtle, ntriples, and csv through the --format option.
The turtle and ntriples formats are RDF 1.2 Message Logs. Every Connection is serialized as one RDF Message, with its quads kept together behind @message . or MESSAGE delimiters. The output uses RDF/JS terms and rdf-parser-ts for serialization and parsing.
For big GTFS files, your memory may not be sufficient. Luckily, we’ve implemented a way to use your hard disk instead of your RAM. You can enable this with an option: gtfs2lc /path/to/extracted/gtfs -f json --store LevelStore.
Conversion uses up to eight workers by default. Use --workers <count> to set an explicit limit.
It may also be the case that your disk has limited storage space. In that case you may want to use the --compressed option.
When you download a new GTFS file, all identifiers in there might change and conflict with your previous export. Therefore, we need to think about a way to create global identifiers for the connections, trips, routes and stops in our system. As we are publishing our data on the Web, we will also use Web addresses for these global identifiers.
See baseUris-example.json for an example on URI templates of what a stable identifier strategy could look like. Copy it and edit it to your likings. For a more detailed explanation of how to use the URI templates see the description at our GTFS-RT2LC tool, which uses the same strategy.
Resolver expressions support property access from connection, trips, routes, and stops, plus format(...) and substring(...). Arbitrary JavaScript is intentionally rejected.
Now you can generate Linked Data in JSON-LD as follows:
gtfs2lc /path/to/extracted/gtfs -f jsonld -b baseUris.jsonThat’s it! Want to serve your Linked Connections over HTTP? Take a look at our work over here: The Linked Connection’s server (WIP)
In GTFS, joining and splitting trains are fixed in a horrible way. See https://support.google.com/transitpartners/answer/7084064?hl=en for more details.
In Linked Connections, we can solve this gracefully by adding a nextConnection array to every connection. A splitting train is then, on the last connection before it is split, indicate 2 nextConnection items.
On your newline delimited jsonld file, you can perform this script in order to make that work: linkedconnections-joinandsort yourconnectionsfile.nldjsonld
Next to the jsonld format, we’ve also implement the “mongold” format. It can be directly used by the command mongoimport as follows:
gtfs2lc /path/to/extracted/gtfs -f mongold -b baseUris.json | mongoimport -c myconnectionsMind that only MongoDB starting version 2.6 is supported and mind that it doesn’t work at this moment well together with the post-processing step of joining trips.
For more options, check gtfs2lc --help
Install the pinned dependencies with npm ci, then run:
npm run check
npm testnpm run check runs ESLint, Prettier verification, and JavaScript type checking. Tests run against an isolated copy of the sample feed.
Run npm run build:sncb to download the latest SNCB/NMBS GTFS feed and generate Jelly-RDF connections for seven service days, starting today in the Europe/Brussels timezone. Results are written to build/sncb/output.
The build can be reproduced or customized with environment variables:
SNCB_START_DATE=2026-06-28 SNCB_DAYS=7 SNCB_FORMAT=jelly npm run build:sncbSNCB_BUILD_DIR, SNCB_GTFS_URL, SNCB_TIMEZONE, SNCB_WORKERS, and SNCB_COMPRESSED=1 are also supported.
gtfs2lc-build searches the Mobility Database API, lets you select a GTFS Schedule feed, discovers its publisher or hosted ZIP URL, and builds the dataset. For example:
MOBILITYDATABASE_REFRESH_TOKEN=... gtfs2lc-build "De Lijn"The first run interactively configures stable public URI templates. It explains the collision risks, recommends agency-scoped identifiers, includes service dates in trip identifiers, includes stop_sequence in Connection identifiers, and previews representative identifiers before saving. The resulting file is named from the agency and Mobility Database key—for example, de-lijn-mdb-684.json—and can also be passed directly to gtfs2lc --base-uris.
Later runs reuse saved configurations without querying Mobility Database:
gtfs2lc-build mdb-684The API accepts MOBILITYDATABASE_ACCESS_TOKEN directly or exchanges MOBILITYDATABASE_REFRESH_TOKEN for a one-hour access token. If neither is set, the command securely prompts for the refresh token and never saves it. Use --help for output, format, compression, worker, and configuration-directory options.
We first convert stop_times.txt to connection rules called connections.txt.
Service dates are processed through calendar_dates.txt and calendar.txt, that was processed at the same time.
In the final step, the connection rules are expanded towards connections by joining the days, service ids and connectionRules.
Post-processing steps work directly on the output stream, and can map the output stream to Linked Data. Connections2JSONLD is the main class to look at.
Another post-processing step is introduced to fix joining and splitting trips.
At this moment we've only implemented a conversion from the Stop Times to connections. However, in future work we will also implement a system for describing trips and routes, a system for transit stops and a system for transfers in Linked Data.
Furthermore, also frequencies.txt is not supported at this time. We hope to support this in the future though.
-
Pieter Colpaert [email protected]
-
Julián Rojas [email protected]