Runs vaultwarden on a GitHub Actions runner, reachable over Tailscale. No dedicated server — just recurring jobs.
.github/workflows/vaultwarden.ymlstartsvaultwarden/server:latestin Docker on anubuntu-latestrunner.- Tailscale joins the runner to your tailnet as hostname
vault(--ephemeral, so the node auto-removes on disconnect), reachable atvault.<tailnet>.ts.net. - The Tailscale TLS cert is cached in GitHub Actions cache and only renewed when it's missing or expiring within 30 days — avoids Let's Encrypt's duplicate-cert rate limit (5/week per domain).
- A loop checks the container every 60s and restarts it if it dies.
- GitHub-hosted jobs cap out at 6h, so a
schedulecron fires a fresh run every 5h.concurrency: cancel-in-progress: truekills the old run once the new one starts — only one instance up at a time.
- Vaultwarden's database is external Postgres (Supabase) — survives every restart.
- Vaultwarden's local
/data(attachments, RSA keys) is not persisted — it lives in the ephemeral runner and is wiped every ~5h restart. Fine for password data via sync; don't rely on file attachments/sends.
Set repo secrets:
gh secret set TS_AUTHKEY -b"<tailscale auth key>"
gh secret set TS_API_KEY -b"<tailscale API access token>"
gh secret set TS_TAILNET -b"<tailnet name>"
gh secret set TS_CERT_DOMAIN -b"<vaultwarden.<tailnet>.ts.net>"
gh secret set VAULTWARDEN_DB_URL -b"<postgres connection string>"
Trigger manually via Actions tab (workflow_dispatch) or wait for the next scheduled run.
- No persistent
/datavolume — attachments/sends lost on restart. vaultwarden/server:latestis unpinned — can break on upstream image changes.- Let's Encrypt allows only 5 certs/week for the same domain — the Actions cache avoids this, but if the cache is evicted (10 GB repo limit / 7-day no-access eviction) and a renewal is needed, the run may fail until the limit refills (1 per 34h).
- This is a workaround, not a real deployment. A small VPS with a persistent volume is more correct if uptime/data durability actually matter.