Skip to content

Updates from the admin UI

Sellf checks releases in every deployment. Settings → System and the dashboard notification offer an in-place update only for a ready standalone PM2 installation. Other deployments display the new version and instructions for their deployment. The upgrade API enforces the same capability as the UI.

Set the server environment variable SELLF_DEPLOYMENT_RUNTIME to pm2, systemd, docker, compose, coolify, vercel, netlify or manual. An unknown value selects manual instructions. An override selects the instructions; it cannot turn an incomplete or container installation into an in-place updater.

Without an override Sellf detects Vercel (VERCEL=1), Netlify (NETLIFY=true), Coolify (COOLIFY_RESOURCE_UUID / COOLIFY_CONTAINER_NAME), container markers (/.dockerenv / /run/.containerenv), PM2 (pm_id), then systemd (INVOCATION_ID). An unrecognised deployment uses manual instructions. Compose uses compose by default in the supplied compose file. Set SELLF_DEPLOYMENT_RUNTIME=coolify on Coolify resources, including Dockerfile and Compose deployments.

In-place mode requires Linux, root, an active systemd installation, an executable /usr/bin/systemd-run, a writable standalone tarball installation, its updater and .env.local, and the matching PM2 process in /root/.pm2. The process name must match the stack directory containing admin-panel. The supported working directory is <stack>/admin-panel/.next/standalone/admin-panel. StackPilot’s standard PM2 installation uses this layout. Non-root, source-based, clustered or systemd-only installs use their existing deployment procedure instead.

Back up the database, check out the target release, and apply its migrations before starting the new app:

Terminal window
bunx supabase db push --db-url '<your-database-connection-url>'

Containers, Coolify, Vercel and Netlify do not apply migrations on startup. The image contains SQL files for reference, but its startup command only starts Node. Supabase is a separate service, including in the full-stack deployment.

Runtime Application update
StackPilot standalone PM2 Use the UI when it offers the update action. The signed updater applies migrations through the service-role RPC, replaces the persisted build, restarts PM2 and saves its configuration.
Plain PM2 from source Check out the release; cd admin-panel; bun install --frozen-lockfile; bun run build; pm2 reload ecosystem.config.js (single process: pm2 restart ecosystem.config.js). Use your existing ecosystem file.
systemd only Build the release as above, then sudo systemctl restart <your-sellf-service> using your unit’s working directory and service name.
docker run Pull ghcr.io/jurczykpawel/sellf:<version> and recreate the container with the new image, preserving environment, ports and networks. Restart alone retains the old image.
Compose: docker-simple / full-stack Set SELLF_VERSION=<version> in .env; docker compose pull sellf; docker compose up -d --force-recreate sellf.
Coolify Set the image tag, or the release commit for a Dockerfile build, and Redeploy. For a Compose resource, set SELLF_VERSION and redeploy the resource.
Vercel Deploy the release source with the existing project environment; wait for Ready and promote to production.
Netlify Deploy the release source with the existing site environment; wait for Published.

After deploying, use Check for updates to verify the running version. The UI verifies the live version after an in-place restart and retains script failures. An unavailable launcher is returned as a failed API request, not an accepted update. Inspect the instance upgrade log when an in-place update fails.

Updater 2 authenticates the manifest with the installed release key, checks the signed min_updater_version, validates the archive and extracts into a private temporary directory. It then executes the downloaded release’s scripts/upgrade.sh (stage 2) with the same token, installation, progress/log paths and instance lock. The panel shows the same steps. Packages without stage 2 use the installed fallback. Migration failures restore old code before it starts; committed database migrations are not reversed (the RPC transaction is still per file).

2026.11.0 requires updater 1 (the unversioned 10.6 script). CI publishes a signed three-line V2 manifest alongside the legacy two-line manifest for that bridge. The old script cannot delegate on its first hop, and its self-copy may leave a trailing shell error even after reporting success. Use an up-to-date Stackpilot checkout to use the downloaded runner during that first hop. If the panel reports an old updater or cannot verify the signed requirement, retry the check or run from your local Stackpilot checkout, replacing <server> with your SSH alias:

Terminal window
./local/deploy.sh sellf --ssh=<server> --update

SELLF_UPDATE_CHECK_URL optionally points the server to a release metadata mirror or local test fixture, using the GitHub release JSON format. The default is GitHub’s latest-release API. The request has a five-second timeout and does not follow redirects. This setting only changes notification metadata: upgrade.sh still fetches official release assets and verifies the installed signing key. Do not configure a mirror advertising a release absent from GitHub.

Run cd admin-panel && bun run test:upgrade-matrix. See the suite README for prerequisites, disposable ports, evidence and remaining platform limits.