Over the past fortnight, 13 news organizations published 18 new open-source repositories — a run of data projects and trackers alongside a notable cluster of tools built for, and with, AI coding agents. Highlights below.
Highlights
After temperatures topped 115 degrees across eastern Montana on Sunday, July 12, Montana Free Press released the statewide temperature analysis behind its July 16 story “Eastern Montana shatters heat records.” Jacob Olness‘s scripts pull every Montana station in NOAA’s GHCN-Daily archive — including discontinued historical ones — drop observations that carry a quality flag, and rank every daily maximum on record so the weekend’s readings can be measured against more than a century of data.
Reuters Graphics shipped a bunch of new tools. The centerpiece is Spectre, a cross-browser UI-audit harness that runs a Playwright sweep across a project’s routes on emulated phones, tablets and desktops, capturing console errors, failed requests, accessibility violations and screenshots into a browsable report; it also emits an LLM triage prompt that clusters the findings by severity for a coding agent to work through. Alongside it came a package of watercolor-style country flags with ISO-code lookup, a set of TopoJSON country and border assets for its mapping client, and a diagnostic plugin that flags CSS classes wrongly dropped by build-time purging.
Canada’s The Globe and Mail open-sourced the code behind its wildfire tracker, which Murat Yükselir built to pull active-fire statistics every hour from the Canadian Wildland Fire Information System and refresh a Datawrapper chart of fire counts and hectares burned, broken down by containment status. The charts run in the paper’s ongoing wildfire coverage.
Northwestern’s Knight Lab built mynews, an experiment in giving readers more control over their news diet. Members of a small, invite-only group subscribe to RSS feeds and get an algorithmically ranked “best of” rather than a reverse-chronological dump — but every article shows why it surfaced, and both sliders and “more/less like this” buttons tune the ranking out in the open instead of through a black box. Readers can also peek at another member’s feed to see past their own filter bubble. The repo is also unusually legible about its own making: it ships a plain-language design brief laying out the product and a dependency-ordered, ten-phase project plan — spec-grade documents that read as much for a coding agent as for a human team.
The Minnesota Star Tribune opened a configurable precinct-level election map embedder for Minnesota’s 2026 primary. Jeffrey Hargarten‘s Svelte-and-Mapbox tool lets newsroom staff generate embeddable results maps for a given office through a configuration page — no code required — and download the underlying GeoJSON and result CSVs for other reporting.
Norway’s NRK released a small MCP server that makes Open Knowledge Format (OKF) knowledge bases searchable by AI agents. It indexes each file’s YAML frontmatter and exposes tools to list tags, list documents and fetch one, so an assistant can retrieve the right doc without needing to crawl the file tree. I’m a big fan of LLM knowledge-bases, though I use the Karpathy LLM Wiki format.
By the Numbers
Beyond new repos, 84 news organizations made a combined 4,821 public commits to GitHub during this period. The most active by commit count (excluding, as best we can, commits done by bots, gh-actions, or cron):
| Organization | Commits |
|---|---|
| The Guardian | 1,363 |
| Freedom of the Press Foundation | 587 |
| ICIJ | 276 |
| OpenSanctions | 256 |
| OpenElections | 170 |
| MuckRock | 162 |
| Reuters Graphics | 158 |
| The Pudding | 105 |
| Vox Media | 95 |
| OpenAleph | 94 |
This roundup is assembled with the help of open-journalism-bot, which monitors public GitHub activity from ~420 news organizations and posts new repositories to @openjournalism.news on BlueSky. Commit counts shown here exclude commits we identified as automated.