Blog
Insights, analysis, and updates from the AI agent economy. Browse by tag · Browse the archive.
SGLang Crossed 400,000 Production GPUs: Picking Your Self-Hosted PaaS's Default Model Server
SGLang now serves 400,000+ production GPUs, making it a real second default next to vLLM. Benchmarks, workload maps, and a decision matrix for choosing the model server on your own GPU pool.
Sevalla Courts Heroku Refugees With $5 Pods — but Price the Database First
The same web-plus-worker-plus-Postgres stack priced on Heroku, Sevalla, and a Hetzner box: ~$120 vs ~$113–$223 vs ~$10 a month — and why the database tier, not the $5 pod, decides the migration.
The 2026 Self-Host-to-Managed Boomerang: Why Day-2 Ops Burnout Sends Teams Back to Render and Railway at 3-5x the Cost
Teams that fled Heroku for Hetzner are quietly moving back to Render and Railway. The documented retreats, the real 3-5x math including ops labor, and the runbook that decides who stays self-hosted.
Red Hat Ships MCP Servers for OpenShift and Ansible: The Enterprise Read on Agent-Operated Infrastructure
Red Hat's MCP servers for OpenShift and Ansible went GA in May 2026 with a read-mostly, playbook-gated design — a reference for agent-operated infrastructure that no git-push PaaS has matched.
Railway Will Forward Your Email Now: What Platform-Owned Domain Perks Really Cost Portability
Railway's September 11 changelog adds email forwarding for Railway-bought domains — one more capability that only exists while the platform holds your registration, DNS, and mail records. A link-by-link accounting of the exit chain, and the portable setup that avoids it.
One Developer, Many Side Projects, One Railway Bill: Where Per-Service Metering Loses to a Flat Hetzner Box
A real Railway bill for 16 idle resources anchors a worked N=1/5/10/50 cost comparison — metered vs fixed-floor vs one flat Hetzner box — with the sprawl crossover named per pricing philosophy.
Hosted vs BYOC vs Self-Hosted: Price Your PaaS Exit in Three Columns
Qovery's September 2026 comparison prices hosted, BYOC, and self-hosted PaaS side by side from vendors' own pages. We turn it into an exit spreadsheet for one typical workload — hosted in the tens of dollars, a control-plane fee that dominates BYOC at small scale, flat hardware plus honest ops hours self-hosted — and map the scale points where the ranking flips.
Qovery's Heroku-Alternative Price Table, Recomputed: What Render, Railway, and Fly.io Really Cost for One Small App
Qovery dated every entry price in its September 2026 Heroku-alternatives comparison. This post recomputes the same small production stack on Render, Railway, Fly.io, and a flat Hetzner box, and shows where usage metering flips the ranking.
Read-Only by Default: The 5-Rung Agent Permissioning Ladder Every Deploy Platform Should Copy
Qovery's agent stack splits deploys from ops and ships its MCP server read-only — with a CLI allowlist and Rego policy tokens above and below. How the five-rung ladder works, where agent-side enforcement breaks, and the checklist for deploy-from-chat on machines you own.
Subscribe
New posts land in your reader as soon as they publish. Pick a format — all three carry the same posts.
Current feeds keep roughly two days of posts so daily polling does not miss a burst. Older entries stay reachable from the feed's next-page link in readers that follow it, or from the blog archive.
Following one topic instead? Browse tags