1732 posts tagged with "Self-Hosting"
Running your own PaaS and infrastructure on machines you own
Railway's Observability Glow-Up: What Six New Panels Are Worth Against a Grafana You Own
Railway's July 31 changelog adds six observability panels with synced cursors and drag-to-zoom — free and good until 30-day retention, zero custom metrics, and click-ops dashboards bind. A worked sidecar cost comparison against Grafana and Perses on owned iron, and the OpenTelemetry move that keeps telemetry portable.
Railway Built an Agent-Execution Cloud in Six Changelogs — Here's the Open Stack That Replicates It
Railway's June–July 2026 changelogs assembled sandboxes, Docker-in-sandbox, checkpoints, preinstalled agents, and chat steering into one closed agent-execution cloud. A primitive-by-primitive teardown plus the open equivalents — Firecracker microVMs, Kata, snapshots, and a git-push PaaS — that replicate it on machines you own.
The Same 2 vCPU, Priced 7x Apart: What Agent-Sandbox Pricing Reveals About Isolation Cost
Seven agent sandboxes price an identical 2 vCPU/4GB hour from $0.052 to $0.378 — a 7.3x spread both of whose extremes are KVM. Duty-cycle math, pause semantics, and owned-hardware breakevens that show what the premium really buys.
Whose Fingerprint Is on This Deploy? Separate Kubernetes Identities for Humans, CI, and AI Agents
A shared deploy token cannot tell a human retry from a CI job or an MCP agent's mistake. Here is the three-lane identity design — SSO-backed humans, OIDC-minted CI, per-agent ServiceAccounts — that makes every deploy attributable without breaking git-push.
7,000 Integrations at a Penny a Call: What a Hosted MCP Registry Costs vs Owning the Agent Path
Arcade.dev's hosted MCP registry charges $0.01 per tool call — about $345 a month for one busy ops agent. This cost-and-control teardown compares renting 7,000 integrations against running your own MCP server on an open platform API.
Buildpacks Aren't Dead: Why Git Push Still Beats a Hand-Written Dockerfile for Most Apps
Dokploy's 35,000 stars celebrate a compose-first deploy contract, but most deploys are a single app with no dependency graph — a grounded ruling on where compose-first genuinely wins and why detection-first still owns the common case.
Cilium LB-IPAM and BGP on Bare Metal: What Dropping MetalLB Looks Like on a Hetzner Fleet
MetalLB 0.16's FRR-K8s default is dead weight on L2-only Hetzner clusters. This post compares staying on MetalLB versus Cilium LB-IPAM with L2 or BGP advertisement, sketches a phased per-cluster migration, and names the catch: on Hetzner dedicated, the BGP peer is a router you build yourself.
Cloud Native Buildpacks Graduates CNCF: What It Means for Teams That Stopped Writing Dockerfiles
Cloud Native Buildpacks graduated CNCF on August 11, 2026 with 535 contributors and a public security audit — what the stamp changes for git-push platforms betting on pack build, and the builder bus-factor warning it doesn't cover.
Your AI Agent Can Restart Production Now: What a Coolify MCP Bridge Teaches About Infrastructure Context vs Code Context
A community MCP server plugs an AnythingLLM agent into the Coolify API for one-prompt restarts and log diagnosis; this case study maps the five gaps between that demo wiring and a secure agent-to-infrastructure bridge, plus the six controls that close them.