Ecosystem

Mojourney, a census of Mojo on GitHub

8 November 2026 · mojourney.dev

Mojourney is a census of the public GitHub repos that contain Mojo. It counts them, reads their git history, and shows how the ecosystem grows and which Mojo versions it moves to. It is an independent project. Mojo is a trademark of Modular Inc., and the census is not affiliated with Modular.

At the time of writing, the overview says:

	now	30 days ago
repos found	1,167
repos with their history read	564	545
commits	103,921	94,677
lines of Mojo	19.0 M	16.5 M

That is 9,244 commits in the last 30 days, a little fewer than the 10,163 in the 30 days before. 528 of the 564 repos read had a commit in the last year. More repos are found than read: the history of the repos found most recently is still being read, a few dozen at a time. The site refreshes every 30 minutes, so these numbers will have moved by the time you read this.

How repos are found

GH Archive, by Ilya Grigorik, records every public GitHub event, an hour at a time, and has done so for years. The census reads every hour since 2023-01-01 and keeps the events that hint at Mojo. It then asks the GitHub API about each candidate once, and keeps the repos whose language or manifests show Mojo.

For each repo it keeps, it clones the repository and reads its first-parent history. At each commit it records the Mojo files and lines, the tags, and the Mojo version the lock file pins. A change of that version between two commits is a migration. It also reads the manifests at HEAD (pixi.toml, mojoproject.toml, pyproject.toml, recipe.yaml and a few others) for the Mojo version they state and the dependencies they declare.

Because the history comes from git, the growth curves reach back before GH Archive first showed the census a repo. The earliest commits in the census are from April 2023.

What the site shows

The main page has the headline counts, a breakdown of repos by Mojo version, and four charts under Growth: lines of Mojo, repos and commits per month, and the share of repos by tooling. Today 88% of repos Of the repos read, 82% use pixi, 15% have no lock file or manifest, and 1% use Bazel, and more than half, 308 of 564, are on Mojo 1.2.

Below the charts are three tables:

Rising ranks repos by what they gained in the last 30 days. The score is stars gained + 3 × forks gained + 5 × the census repos that use it now.
Most active lists the repos with the most commits in the last 30 days, and how many lines of Mojo they grew by.
Most depended on lists the repos that the most census repos declare as a dependency. An owner’s repos that depend on each other share one row, so magmalake’s own libraries are magmalake/*, used by 23 of its repos and none from outside yet. The most used from another owner is Brian Grenier’s EmberJson, used by 5 repos.

Migrations counts, per month, the repos whose lock file moved to each Mojo version. Releases show up as spikes. In July, 209 repos moved to 1.0.0. In August, 197 moved to the nightly 1.1.0.dev2026081105, and in September 247 moved to 1.2.0.dev2026092605.

Repos is the full list, sortable and filtered by Mojo version, activity and first commit. Each repo has its own page, for example bgreni/EmberJson: lines of Mojo over time, stars per month, what it depends on and what uses it, its tags with the Mojo version at each, and every migration. The search box in the header finds repos by name.

Every chart and table has a small (?) next to its title. Click it for what the numbers mean and how they are counted. How it works describes the pipelines in more detail.

The data is also JSON: /api/repos for every repo and /api/overview for the headline numbers and the monthly series.

How it’s built

Seven Mojo pipelines do the work: gharchive, gharchive_index, mojo_repos, mojo_history, mojo_versions, mojo_activity and mojo_stats. They write Apache Iceberg tables to Hetzner Object Storage with iceberg.mojo and parquet.mojo, both written in Mojo. The code is in morca-dev/census.

The pipelines are run by morca, the orchestrator, also written in Mojo. Its runs are public at orca.mojourney.dev. The next post, on Monday, is about morca.

The site is a Mojo HTTP server built on flare, by Ehsan M. Kermani. It reads the Iceberg tables itself, with no database in between. Every 30 minutes it checks the tables’ snapshots and rebuilds its pages in memory when one has changed. Before a new build goes live it runs a set of data checks, for example that the numbers in the top tables match those on each repo’s page. The latest results are at /checks.

The site, the orchestrator and most of the pipelines run on one Hetzner CAX21, a 4-vCPU ARM server with 8 GB of RAM that costs a fixed €12.49 a month. A second machine, an Apple-silicon Mac, runs the git-heavy pipelines as a morca remote worker.

What to keep in mind

Commit counts are first-parent. They follow the main line of each repo’s history, so they will be lower than the totals GitHub shows.
Stars are refreshed daily, from the GitHub API at 06:00 UTC.
GH Archive is loaded, the histories are catching up. 1,406 of the 1,407 days from 2023-01-01 to 2026-11-07 are loaded; the last, 2025-06-12, is missing an hour GH Archive never published, and loads without it. The backfill found many repos in its last days, so about half of the 1,167 repos don’t have their history read yet, and counts that come from the history (commits, lines, versions) cover the 564 that do. The home page shows both fractions.
Detection can miss repos. If a repo’s events never carry a Mojo-specific signal, the census does not see it. Some repos also have no version it can read. They are listed as no-manifest or unknown.

If a repo is missing or something is counted wrong, the Feedback link in the site’s footer opens an issue on census.

Once every repo’s history is read, the commit, line and version counts will cover all the repos found. After that, the plan is to keep the census current and add views as questions come up.