Blog
Insights, analysis, and updates from the AI agent economy. Browse by tag · Browse the archive.
France's Railway Runs on Cluster API and Talos Now: What SNCF's Zero-Drift Rebuild Teaches a Self-Hosted PaaS
SNCF cut cluster provisioning from a month to 30 minutes with Cluster API and Talos Linux. A look at the numbers behind the rebuild and what self-hosted platform teams should copy.
Fixed vs Metered: What Render, Railway, and Fly.io Really Cost for Three Typical Workloads
Render's fixed instance prices, Railway and Fly.io's per-second metering, and a flat-rate Hetzner box — costed line by line for a steady web service, a bursty cron job, and an idle staging env, with the utilization crossover where the ranking flips.
Render Now Mints Short-Lived OIDC Tokens for Anthropic and OpenAI: The Beginning of the End for LLM API Keys in Env Vars
Render's July 2026 changelog lets services authenticate to Anthropic and OpenAI with short-lived OIDC tokens instead of pasted API keys. How the federation works, the setup on each side, and the native equivalent on self-hosted Kubernetes.
Railway vs Render vs Fly.io in 2026: Auditing the Verdict Table, Then Pricing the Hetzner Row It Never Includes
Techsy's 2026 Railway-vs-Render-vs-Fly.io verdicts mostly survive September — but the fine print doesn't: a wrong idle-pricing claim, a regional egress tripwire, a stale seat price, and a missing Hetzner column that undercuts every managed row.
railway.json vs render.yaml vs fly.toml: What Each PaaS Config File Actually Declares
Railway, Render, and Fly.io each ask for a config file, but only Render's Blueprint declares databases, cron jobs, and preview environments. A clause-by-clause comparison plus the exact render.yaml surface a Render-compatible platform must honor to absorb migrators without repo rewrites.
Railway Edge Rules vs a Gateway You Own: What Each Rule Costs to Replicate
Railway's Edge Rules put allow, block, challenge, redirect, and cache policy in a vendor console with no CLI. A rule-by-rule mapping of what each one costs to rebuild on a gateway you own — plus the migration checklist for policy that lives outside Git.
Railway Shipped Three Agent Tracks at Once: Remote MCP, railway agent, and One-Command Skills — Which Should a Self-Hosted PaaS Copy First?
Railway's April 2026 changelog shipped a 7-tool remote MCP server, a railway agent CLI command, and one-command skills install together. How the three tracks fit — and the copy order a self-hosted PaaS should follow: skills first, MCP second, hosted agent last.
Railpack Is Eating Zero-Config Builds: What Coolify v4.1's Bet Means for Your Self-Hosted Fleet
Coolify v4.1 shipped Railpack as a first-class build option, making it the second platform to bet on Nixpacks' successor. Why the zero-config build layer is consolidating, how Railpack's BuildKit-native pipeline beats Nix on image size and versioning, and what self-hosted fleets should do about it.
The Heroku-to-Kubernetes Exit Industry, Mapped: 4 Vendors, 6 Steps, and an Honest Downtime Budget
Qovery's September 2026 guide maps the Heroku-to-Kubernetes exit market into four buyable vendor categories and one repeatable six-step cutover. The honest version: stateless services move with zero downtime, but promoting a new Postgres primary needs a planned write pause of seconds to minutes.
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