Which branch deploys where, and how each app gets there. It covers Jenkinsfile, .github/workflows/site-publish.yml, .github/workflows/test.yml, .github/workflows/e2e-daily.yml, .github/dependabot.yml, .github/workflows/playwright-reviewers.yml, the per-app vercel.json files (for example apps/web/vercel.json) and docker-compose.prod.yml.
| Branch | Role |
|---|---|
develop |
Default branch and staging. CMS content PRs target it. |
master |
Production. |
deploy-develop |
Built static site for staging, served at dev.logos.co. |
deploy-master |
Built static site for production, served at logos.co. |
Branch protection blocks direct pushes to develop and master. The deploy-* branches are written by the build jobs below, never by hand.
Vercel deploys every pushed branch by default. Both publish jobs run scripts/site-publish/skip-vercel.mts, which adds a vercel.json turning deployments off at each Vercel project's root directory, so deploy-* pushes don't start builds. The docs bot branch docs/repo/sync is skipped through git.deploymentEnabled in each app's vercel.json.
The site is a static export of apps/web/out, built with pnpm turbo run build --filter=web --force. The --force is deliberate: Turbo hashes source files, and most of the media comes from an external CMS, so a cache hit could ship last week's articles. Builds that fetch that media run scripts/capture-press-snapshot.ts first and pass the result through PRESS_CONTENT_SNAPSHOT, so feeds, cards and article pages all use one fetch.
Every build takes the origin for canonical links, JSON-LD, robots and sitemap URLs from canonicalUrl in content/site/en/settings.json (https://logos.co). When VERCEL_ENV is preview and VERCEL_URL is set, the build uses https:// plus the VERCEL_URL hostname for Open Graph page URLs and for Open Graph and X card images hosted on the canonical origin, so new share images work before production deploys. Card images hosted elsewhere, such as the CMS, keep their own URL. Other builds use the canonical origin for those tags. Canonical links and indexing rules are the same on every build. Keep Vercel's system environment variables enabled. A link preview service must be able to fetch the page and image without signing in.
A failed fetch from the blog CMS or the active circles API fails the build by default. Set ALLOW_ACTIVE_CIRCLES_FETCH_FAILURE=1 to allow only Active Circles to reuse its committed snapshot on error. The release test workflow sets this option. The Active Circles page shows the date of the snapshot and excludes past events from its upcoming count.
Set ALLOW_CONTENT_FETCH_FAILURE=1 to fall back to the snapshots committed in apps/web/lib/data/ instead: press.snapshot.json for /media and active-circles.snapshot.json for the circles pages. Both are rewritten by every build that reaches its API (generate:media-assets and sync:active-circles), so commit them from a local build now and then to keep the fallback recent. In that mode the build skips image downloads and reuses the resized images in the Vercel build cache; images missing from it keep their CMS URL and stay broken until the CMS is back. Redeploy with the build cache enabled. The CMS requests time out after 20 seconds. Once the CMS proves unreachable (no key, a dropped or timed out request, a 5xx), the rest of that process reads the snapshot directly; other CMS errors fall back for the failed read only. A post the CMS listed but the snapshot lacks is left out of listings and its page renders the 404 page.
During a CMS outage, a preview can also use a previous press-snapshot workflow artifact through PRESS_CONTENT_SNAPSHOT. Check its date and draft status, reuse the available media image cache, then build with vercel build --target preview and upload with vercel deploy --prebuilt. Record the snapshot date in the PR because its media content stays frozen at that point. Homepage statistics requests time out after ten seconds and use their existing fallback values.
Vercel previews for web/muster and web/railgun use the published media snapshot from 5 November 2026 while the CMS is unavailable. Branch-specific PRESS_PREVIEW_SNAPSHOT_URL and PRESS_CONTENT_SNAPSHOT variables supply the archive and its extracted path. The install command downloads the archive only when its URL is set.
generate:media-assets keeps the resized images in apps/web/.next/cache/media-images, which Vercel keeps between builds. An image download fails after 10 seconds without data or 60 seconds in total, with no retry, and the image keeps its CMS URL. The cache then marks that URL as failed for 24 hours, so later builds skip it without a download; vector and animated files are marked as skipped for good. Delete the cache directory, or redeploy without the build cache, to retry them sooner. Jenkins starts from a clean workspace, so it pays the timeout on every run. The active circles requests time out after 10 seconds.
Two jobs can build and publish it.
Jenkinsfile builds the web app with STRAPI_API_KEY and SIMPLECAST_ACCESS_TOKEN, because the /media pages come from the blog CMS and Simplecast, then pushes apps/web/out to deploy-develop or deploy-master with ghp-import. Unless it runs in site publish mode, the same run also builds the CMS image (see below).
site-publish.yml rebuilds only the static site and leaves the Payload image alone. It runs on a site-publish repository_dispatch (the CMS sends one from its dashboard) or by hand through workflow_dispatch, checks out develop for dev or master for production, builds, and publishes the export to the matching deploy branch with scripts/site-publish/publish.mts. Jenkins has the same mode behind its SITE_PUBLISH parameter.
Successful staging builds save the media they fetched as a press-snapshot.json artifact, kept 14 days in the workflow. A production publish names the latest successful staging run (previewBuild in the workflow, PREVIEW_BUILD in Jenkins), reuses its snapshot and checks the site-publish.json marker in deploy-develop before it starts. That way logos.co shows the same media you reviewed on dev.logos.co, even if someone edited Strapi in between. An expired artifact means building a new staging preview.
apps/web/vercel.json sets the build command for Vercel preview deployments.
The web, API and CMS vercel.json files use scripts/vercel-ignore-build.mjs as the ignored build step. It skips the build when nothing the app builds from changed since the last deployment of the branch, or since develop for a new branch. That covers the app, its workspace dependencies, packages/config, content/ for apps using @repo/content and the root build files. Changes to docs folders, READMEs and tests are not counted. A redeploy, or a commit equal to the one it is compared with, always builds.
Deployed on Vercel. apps/api/vercel.json sets the build command to pnpm turbo run build --filter=api.
- Staging:
https://logos-web-api-git-develop-status-im-web.vercel.app - Production:
https://logos-web-api.vercel.app
The CMS ships as a Docker image. Jenkinsfile builds apps/cms/Dockerfile and pushes it to harbor.status.im as logos-web/logos-cms. NEXT_PUBLIC_SERVER_URL and NEXT_PUBLIC_WEB_URL are build args, so they are baked into the image: cms.logos.co in production and dev-cms.logos.co in staging.
docker-compose.prod.yml runs that image next to a self-hosted Postgres, and the container entrypoint applies SQL migrations on boot. Setup details are in apps/cms/docs/deployment.md.
There is also an apps/cms/vercel.json (region iad1, hourly cron on /api/health), but the production path in this repo is the Docker image.
Deployed on Vercel. apps/logos-demos/vercel.json sets the build command to pnpm turbo run build --filter=logos-demos.
Runs locally or in Docker with apps/logos-crm/compose.yaml and apps/logos-crm/Dockerfile. This repo has no Jenkins or Vercel config for it, so where it runs in production isn't recorded here. See apps/logos-crm/docs/deployment.md.
.github/workflows/test.yml runs pnpm test:release on every pull request and on pushes to develop and master. It starts a Postgres 17 service and sets GITHUB_TOKEN for the build-time RFP fetch. For media, it passes the committed press snapshot as PRESS_CONTENT_SNAPSHOT and two representative covers as PRESS_MEDIA_IMAGE_FIXTURES_DIR. The normal image generator resizes these covers, and the static checks still verify exported assets and search links. This fixture directory is used only by CI.
.github/workflows/e2e-daily.yml runs the Playwright suite in e2e/ every day at 05:47 UTC against dev.logos.co (staging) and logos.co (production), in both Chromium and Firefox, checking every sitemap entry. The matrix gives four jobs, one per site and browser; each job installs only its browser and runs pnpm test:e2e --project=<browser>. You can also start it by hand through workflow_dispatch. It also runs on pull requests that change e2e/package.json or the workflow file itself, so a Playwright upgrade is tested against both sites before merge. On the default branch, a failed run opens or comments on the issue labelled e2e-daily, and the next green run closes it; runs on other branches and on pull requests leave the issue alone. Each ref gets its own concurrency group (e2e-daily-${{ github.ref }}), so pull request runs do not queue behind the daily run. Each job keeps its HTML report and traces as the e2e-daily-<environment>-<browser> run artifact for 14 days. The workflow reads the optional E2E_BYPASS_TOKEN secret and sends it as the x-e2e-bypass header, so the server can exempt the run from bot protection and rate limits. It also passes the workflow's GITHUB_TOKEN to the tests, so the Basecamp test can look up the latest release through the authenticated GitHub API: anonymous requests to github.com time out when GitHub is under load.
.github/dependabot.yml checks npm every day at 04:00 for new releases of @playwright/test and playwright, and groups them (with playwright-core) into one PR with the chore(e2e) commit prefix. It skips releases younger than 3 days, which also keeps the lockfile update clear of pnpm 11's 1-day minimumReleaseAge. The suite runs on Playwright's own Chromium and Firefox builds, so these PRs keep both browsers current. .github/workflows/playwright-reviewers.yml requests reviews on these PRs from the logins in its REVIEWERS variable, since Dependabot no longer has a reviewers option. Merge them by hand once the E2E run passes. Security updates for other dependencies are unaffected. Dependabot PRs only read Dependabot secrets, so E2E_BYPASS_TOKEN needs a copy there too.