Hosting, edge compute, databases, and the pipes that hold modern apps together.
75 articles
Workers KV's documented 60-second window is a cache TTL you cannot lower, not a replication delay. Where it bites: cached nulls, uniqueness checks, multi-key updates — and when a Durable Object is the right swap.
CDP gives a scheduled agent a real browser when a site ships no API. Here is what it costs in memory and wall clock, the four failure modes that only surface on a cron, and the cases where you should not do it at all.
A deploy API returning 200 does not mean your new URLs are reachable at the edge. Here is how the deploy-then-ping race silently burns IndexNow submissions, and the verification sequence that stops it.
How to audit an S3-compatible bucket, write lifecycle rules that actually pay for themselves, and avoid the transition fees, minimum-duration charges, and versioning traps that make bills go up instead of down.
k3s, Talos Linux, and a managed control plane solve different halves of the same problem. What each one actually costs you in money, upgrade work, and 2am debugging on a three-node cluster.
Postgres forks a process per connection, so pooling is not optional past a few dozen clients. How PgBouncer, Supavisor, and driver-side pools differ, and what transaction mode breaks.
Blue-green deployments do not require Kubernetes, a service mesh, or a dedicated platform team. Here is a working setup using nothing more than a reverse proxy, two ports, and a shell script.
Edge caching is not just for static assets. A properly configured CDN can serve API responses, authenticated content, and even dynamic pages from a point-of-presence 20 milliseconds from your user — if you get the Cache-Control headers right.
Most backup scripts succeed at creating files and fail at restoring them. Here is how to build a backup pipeline that survives an actual disaster drill — scheduled restores, WAL archiving, and the three things your backup must prove it can do.
You do not need a Terraform monorepo, a dedicated infrastructure engineer, or a complex CI pipeline to get the benefits of infrastructure as code. A single main.tf file, a state backend, and a GitHub Actions workflow that runs on push is enough.
Serverless pricing lowers the barrier to launch, but above a certain traffic volume, per-request billing flips from saving you money to costing you multiples of a $6 VPS. Here is the crossover math.
A practical comparison of Caddy and Nginx for solo developers and small teams: certificate management, performance trade-offs, config ergonomics, and when switching actually pays off.
A measured comparison of Hetzner and OVHcloud for side projects in 2026 — pricing models, bandwidth, hardware, and the trade-offs that actually matter for a solo developer.
A measured look at what swapping Node.js for Bun changes in real production: install speed, native TypeScript, built-in tooling, and the compatibility and observability gaps that still bite.
A practical comparison of Coolify and Dokploy for solo devs running their own deployments in 2026 — architecture, setup, resource use, and which one to pick.
A developer's comparison of Turso (edge SQLite/libSQL) and Neon (serverless Postgres) in 2026 — latency model, branching, cost shape, and lock-in.
A measured look at Kamal 2 — 37signals' Docker-over-SSH deploy tool. What kamal-proxy changed, where it fits, and where it hands the hard problems back to you.
We migrated three production stacks across Caddy 2.8, Traefik v3.1, and nginx Proxy Manager 2.11. Here is where each one earns its keep and where it bites you.
Three lightweight Kubernetes distributions — K3s, MicroK8s, and k0s — compared on memory footprint, default add-ons, HA story, and which team shape each one actually fits. Operational opinions, not synthetic benchmarks.
We deployed the same three-service app to Coolify, Dokku, and CapRover on a $12/mo Hetzner box. Here's the architecture, memory footprint, and ops cost of each self-hosted PaaS in 2026.
Block transferred its Goose agentic AI framework to the Linux Foundation. Here's what vendor-neutral governance means for teams choosing between LangChain, AutoGen, and Goose — and the lock-in risk most teams overlook.
We deployed a Go API and Next.js app across Fly.io's edge network, measuring cold starts, regional latency, and DX against Railway, Render, and Heroku — plus WireGuard networking and fly.toml deep-dive.
We migrated 18 Docker containers from Docker Desktop to OrbStack on an M1 Max MacBook Pro — measuring memory, CPU idle, and cold starts. Review of macOS-native architecture, Docker API compat, and real-world dev performance.
We built payment processing, user onboarding, and AI orchestration on Temporal — measuring durability, replay, and SDK learning curve vs Step Functions and job queues. Review of workflow-as-code, deterministic execution, and where durable execution replaces retry logic.
We integrated Turso's libSQL SDK into a TypeScript analytics pipeline with embedded replicas across 3 regions — review of the architecture, replication model, and how it compares to Cloudflare D1, PlanetScale, and vanilla SQLite.
We replaced self-hosted Redis and Kafka with Upstash's serverless offerings, measuring latency from 3 regions vs AWS ElastiCache and Confluent Cloud. Review of Redis REST API, Kafka HTTP bridge, and where per-request pricing wins.
A close read of NVIDIA CUTLASS — the header-only CUDA template library behind a surprising amount of modern AI infrastructure. What it is, how its kernel hierarchy works, where CuTe and the Python DSL fit, and when to reach for it.
A measured look at where AMD ROCm with PyTorch and PyTorch Lightning still has rough edges on the RX 7900 XTX in 2026, and what that means if you are porting CUDA training workloads.
OpenAI swapped ChatGPT's default to GPT-5.5 Instant overnight, claiming faster responses, sharper reasoning, and fewer hallucinations. We grade each claim against independent testing and show developers what to change in their API stack.
OpenAI's Daybreak and Anthropic's Glasswing shipped the same week with matching cybersecurity benchmarks and overlapping enterprise partners. Here's what the convergence signals and how to evaluate either for your AppSec pipeline.
Macchiato's Day 2 release ships a live token sidebar, per-agent cost dashboard, and shortcuts for Claude Code and OpenCode. Here is what changes for developers running multiple AI agents.
We installed Coolify on a $20/month Hetzner VPS and deployed an Astro site, Next.js app, and PostgreSQL instance through its dashboard. Here is how the open-source PaaS compares to Vercel and Netlify on cost, reliability, and developer experience.
We tested OrbStack, Colima, and Rancher Desktop as Docker Desktop replacements on macOS and Linux. Measured RAM usage, startup time, and Docker Compose compatibility to find which alternative actually delivers the same development workflow without the licensing headaches.
We compared PlanetScale, Neon, and Supabase on cold start latency, pricing at scale, branching workflows, and ORM compatibility. Here is how the three leading serverless database platforms stack up — and which one fits your application architecture.
SST Ion rewrites the Serverless Stack framework, compiling TypeScript to Terraform and Pulumi outputs. We migrated a production API from SST v2 to Ion and measured cold start performance, deployment speed, and the live Lambda debugger in practice.
We deployed a globally distributed app on Turso's libSQL-based edge database and measured query latency across 5 regions. Here is how its primary-replica architecture performs against Cloudflare D1 and traditional Postgres — and where the SQLite limitations actually matter.
Cloudflare D1 brings transactional SQLite to the edge with zero cold starts and read replication across Cloudflare's global network. A detailed review of the architecture, the Workers integration, and where D1 fits in the serverless database landscape.
Fly.io and Railway both promise to eliminate infrastructure complexity, but they approach deployment from fundamentally different philosophies. A detailed comparison of regions, pricing, developer experience, and which workloads each platform handles best.
Neon rearchitects PostgreSQL to separate storage from compute, offering serverless scaling, database branching, and a generous free tier. A hands-on look at where it shines and where traditional Postgres still wins.
SST Ion reimagines infrastructure-as-code by embedding AWS resource definitions directly into application code, with live Lambda debugging and a Terraform-compatible deployment engine. A review of the developer experience, the Pulumi migration, and where Ion fits in 2026.
Supabase Edge Functions bring Deno-powered serverless compute to the Supabase ecosystem, with tight integration into the database, auth, and storage layers. A review of the runtime, the developer experience, and where it fits versus Cloudflare Workers and AWS Lambda.
Why Promise.race leaks model calls and billing in AI agents, and how a single-owner pattern with AbortSignal, deadline budgets, and jittered retries fixes it.
Temporal's durable execution engine crossed 3,000 paying customers as teams building long-running LLM agents swap DIY retry code for crash-proof workflows. We break down what durable execution buys you and where it costs you.
MinIO's MemKV offloads transformer KV cache to persistent memory tiers so agentic AI pipelines reload attention state instead of recomputing it. We break down the recompute tax, MinIO's 95% utilization claim, and when reload actually beats recompute.
AI agents return 200s and exit cleanly while hallucinating, degrading under rate limits, and overrunning budgets. Here are the four silent failure modes and a minimal monitor that catches them in production.
HTTP's request-response model was never built for AI agents that run for minutes or hours. Here is why connections drop mid-task and how Ably's durable sessions keep messages, state, and reconnects intact.
Observability platforms now let you ask questions of logs, metrics, and traces in natural language. Here's how the translation layer works, what it genuinely helps with, and where it breaks.
A detailed look at Caddy's automatic TLS, Caddyfile syntax, and reverse proxy setup — and where it falls short compared to Nginx.
How Apple Silicon's unified memory architecture makes the Mac Mini a practical local inference node — benchmarks, real costs, setup with Ollama and MLX, and honest tradeoffs versus cloud GPUs.
How Nix flakes and devShells replace Docker for local dev: what works, where it hurts, and whether the learning curve is worth it for your team.
Python dominates ML development but struggles in production serving. The Rust sidecar pattern splits responsibilities: Python handles models, Rust owns the hot path. Here's the mechanics.
A grounded comparison of SendGrid, Mailgun, and Resend across pricing, developer experience, deliverability, and fit — for developers picking a transactional email API in 2026.
SigNoz unifies logs, traces, and metrics on ClickHouse and OpenTelemetry. Here is what it actually costs, where it shines, and where self-hosting bites back.
Most LLM inference waste is a scheduling problem, not a hardware one. Continuous batching, KV-cache management, speculative decoding, and model routing cut energy and cost per token without new silicon.
Temporal Cloud now lets you run durable workflows on AWS Lambda with zero infrastructure management. Here's what changed, what the tradeoffs are, and whether it fits your workload.
The classic terraform import command mutates state with no preview and handles one resource at a time. Here is how config-driven import blocks, generated configuration, and helper tooling change that calculus.
A technical walkthrough of optimizing matrix multiplication in Swift on Apple Silicon — loop reordering, cache blocking, SIMD, multithreading, and GPU offload — and why matmul throughput sets your LLM training speed.
A review of mikeroyal's Self-Hosting Guide, the GitHub resource for running local LLMs, WireGuard VPNs, Home Assistant, and private cloud on your own hardware — plus where self-hosting saves money and where it doesn't.
PyPI's catalog is growing faster than ever. Here's how the surge affects supply-chain risk, dependency bloat, and what to use when you audit your tree.
A measured look at Immich, the open-source self-hosted photo platform with mobile apps and on-device ML, covering deployment, AI features, and the tradeoffs of running it yourself.
A measured review of Supabase — the open-source Firebase alternative built on dedicated Postgres with auth, storage, realtime, and pgvector. What holds up for AI backends, what doesn't, and where pricing and the realtime engine bite.
Coolify is an open-source PaaS you self-host for around $6/month. We tested its 280+ one-click services and where it beats Vercel/Heroku — and where it doesn't.
We read through the rk3562deb project that converts cheap RK3562 Android tablets into Debian Linux machines. Here's what works, what doesn't, and which dev workflows actually fit.
A new poll shows roughly 70% of Americans don't want AI data centers built nearby. Here's how the resulting permitting drag will hit inference pricing, region availability, and your architecture decisions.
Mozilla told UK regulators VPNs are essential privacy infrastructure, not threats to be neutered. Here's why developers rely on VPNs more than they realize, and what changes if Ofcom listens to the other side.
A developer-focused comparison of domain registrars: API access, DNS management, WHOIS privacy, and honest renewal pricing. Stop overpaying for domains you bought years ago and forgot about.
A hands-on look at running Bun-based JavaScript apps on Cloudflare Workers — cold starts, free tier limits, the node:* compat story, and when Workers beats a VPS for developer side projects.
A practical head-to-head between Supabase and Firebase for indie developers building in 2026. Covers Postgres vs Firestore, authentication, realtime subscriptions, pricing cliffs, and when open-source ownership beats vendor convenience.
We deployed the same Next.js app + Postgres database to Fly.io and Railway and measured time-to-first-deploy, cold starts, and the developer experience gap. Railway won on speed; Fly.io won on global reach. Here's the breakdown.
A hands-on review of paperless-ngx, the open-source self-hosted DMS. We cover the Docker stack, OCR pipeline, REST API, AI workflow integration, and where Whoosh search hits its limits.
A hands-on review of Gitleaks for catching hardcoded secrets in Git history. Covers the CLI, pre-commit hooks, CI integration, and how it stacks up against GitGuardian for teams that don't want to pay per developer.
play.pickuma.com now has eleven games. After the first two, the bottleneck was no longer game logic — it was the chrome around each game. Here's the design system that made the next nine fast.
Stop at 7.77 and Eagle Run are live at play.pickuma.com. A 250-line vanilla canvas game and a one-button time-sense test, both shipped in a day. The stack, the tradeoffs, the things that worked.
A malicious Obsidian community plugin delivered the Phantom Pulse RAT to developer vaults. Here is the attack chain and how to audit plugins in Obsidian, VS Code, and Cursor.
A 2026 audit of free-tier developer services: which hosting, database, CI/CD, and observability platforms still let you ship a side project for $0, where the hidden cliffs are, and when paying actually costs less than working around limits.
One email every Friday with the tools worth your time. No spam, unsubscribe anytime.