1512 posts tagged with "PaaS"
Platform-as-a-Service tooling, economics, and self-hosted alternatives
Self-Hosting gpt-oss-20B on a Single 12GB GPU: What OpenAI's Open Weights Cost to Serve vs the API
Serving gpt-oss-20B on a single 12GB GPU costs about $30 a month all-in and breaks even against o3-mini-class API pricing near 15 million tokens — the full VRAM math and worked self-host vs API costing for both model sizes.
Pre-Auth RCE on the Box That Ships Your Code: What TeamCity CVE-2026-63077 Costs and Why Ephemeral Builds Change the Math
TeamCity CVE-2026-63077 gave unauthenticated attackers RCE on every On-Premises version. This teardown inventories the real blast radius — every credential and artifact the build server touches — and shows why ephemeral, narrowly-scoped build containers are a structurally different trust model.
ZaneOps v1.14 Goes Multi-User: Workspaces, Roles, and a Paid License on the Swarm PaaS
ZaneOps v1.14 adds workspaces, four roles with per-project scoping, and an optional Starter license to the MIT-licensed Swarm plus Caddy PaaS — free for 1 workspace and 3 users. This breakdown covers what the multi-user milestone changes, what the paid license actually unlocks, and why the second machine still is not dashboard-managed.
One Compose Command, 30 AI Tools: What selfhost-ai Gets Right — and Where It Hits the Second-Machine Wall
selfhost-ai installs 30+ self-hosted AI tools with one command, proving compose-first distribution won the homelab — but single-host Compose, no tenancy, and lockstep upgrades wall off teams, so here is which pieces a PaaS should absorb as one-click templates.
NETSCOUT Put Its Network Evidence Behind an MCP Server. Your Platform's Telemetry Is Next
NETSCOUT's September 2026 MCP connectivity for Omnis AI Insights sells agents curated network evidence instead of dashboards — plus a concrete checklist for the read-scoped MCP tool surface a self-hosted PaaS should expose over its own metrics, logs, and traces.
Suga's Docker-Lens Heroku Shootout, Audited: What Seven Platforms Really Cost Teams That Ship a Dockerfile
Suga's September 2026 guide profiled seven Heroku alternatives for Docker workloads without price-ordering them. This audit prices the same web + worker + Postgres + Redis stack on all seven plus a flat-rate Hetzner box — and names where each vendor bills container teams worst.
China Sorted Every AI Agent Action Into Three Authority Tiers. Your Deploy Pipeline Should Too
China's July 2026 agent rules require every AI agent action to be classified as fully autonomous, human-reviewed, or human-only before deployment — that sorting applied to a real deploy/operate action list, and why to adopt it before the EU and US follow suit.
The $25 Hetzner Stack: What 8 vCPUs, Coolify, and SQLite-on-NVMe Actually Prove
A January 2026 Show HN runs a full production stack on one $25 Hetzner VPS with Coolify and SQLite — the headroom math holds up, but live file-copy backups don't. The Litestream fix and three priced paths to a second machine.
Your Fleet Is Paying Rent on Disks Nobody Mounts: Reaping Orphaned PVCs With Kubernetes v1.37's Last-Used Timestamp
Kubernetes v1.37's beta PVC Unused condition records exactly when a volume went idle. A seven-stage reaping policy — 30-day discovery, snapshot-before-delete, 14-day review — turns that timestamp into safe, auditable deletion on a self-hosted fleet.