Pages I open on day one of a new project: 8 notes on free-for-dev, public-apis, offline Docker pulls and SQLite at 4M QPS

When I start a new side project, the same handful of pages end up open in my browser before I write any code. Some are giant lists that save hours of searching, a couple are small tools that fix one annoying problem, and one is a benchmark write-up that changed how I think about databases. I keep a short reading note on each in my digital garden, and these eight are the ones I return to most.

Lists worth bookmarking before you build

free-for-dev. This is the list I check before paying for any hosted service. It collects SaaS, PaaS and IaaS offerings that have a real free tier, and it's aimed at people doing infrastructure and DevOps work. The inclusion rules are what make it trustworthy: only hosted services count (no self-hosted software), a free trial doesn't qualify, any time-limited free tier has to last at least a year, and services that lock TLS behind a paid plan are rejected. It covers the big clouds (Google Cloud, AWS, Azure, and Oracle's generous Always Free tier) alongside code hosting, CI, and a long tail of APIs. My notes on the free-for-dev list and its inclusion rules summarize the first part of it.

public-apis. If you need test data or want to wire a quick demo to something real, the public-apis repository is the usual starting point. It's a community-maintained, MIT-licensed collection of free APIs grouped by topic, and with roughly 485k stars it's one of the most-starred repositories on GitHub. I mostly use it to answer "is there a free API for this?" in under a minute. Here's my short entry on public-apis.

awesome-cto. This one is less about code and more about the job around it. It's an opinionated reading list for CTOs and VPs of engineering, tilted toward startups and fast-growing companies. You'll find essays on how the CTO role differs from VP Engineering, how a founder-CTO's job changes over time, hiring and interview loops, engineering management, architecture choices like trunk-based development and microservices, and the business side, including fundraising and technical due diligence. See what's inside the awesome-cto reading list.

SaaS Starter Stack. For anyone trying to ship a small paid product, this repository lists free or cheap building blocks by layer: frameworks (Astro, Next.js, Nuxt, Remix, SvelteKit), boilerplates, databases such as Supabase and Pocketbase, hosting from Vercel to Hetzner and Coolify, payments through Stripe, Lemon Squeezy or Paddle, plus analytics, feedback, live chat and email tools. It also includes interviews with founders whose products make at least $500 in monthly recurring revenue, which is the part I found most useful. I wrote up the SaaS Starter Stack by layer.

Small tools that remove one headache

docker-pull-tar. Moving container images into a machine with no internet access is a chore. This MIT-licensed tool pulls an image straight from a registry and writes a .tar you can carry across, then load with docker load -i. It doesn't need Docker or even Python on the downloading machine if you use the EXE, and it resumes interrupted downloads, checks SHA256 digests, and lets you pick the architecture (amd64, arm64 and others). There's also a Gradio web UI for searching images and choosing tags. More in my note on pulling Docker images offline as tar files.

Setting a proxy in the terminal. I can never remember the exact syntax, so I keep this cheat sheet. It shows how to set, view and clear HTTP and SOCKS proxies in each environment: netsh winhttp on Windows, set and setx in CMD, $env:HTTP_PROXY in PowerShell, export http_proxy and all_proxy in Bash, and git config --global http.proxy for Git. It's handy whenever a package manager or git clone stalls behind a corporate or local proxy. The commands are in my terminal proxy cheat sheet.

Thinking about scale

The System Design Primer (Simplified Chinese edition). Donne Martin's System Design Primer needs little introduction, and its Simplified Chinese README, produced by a volunteer translation project, is a good way to share it with Chinese-speaking teammates. It suggests a study plan by time horizon (breadth first, depth later), treats the interview as an open conversation you should steer, and walks through the trade-offs: latency versus throughput, CAP, consistency and availability patterns, and the building blocks such as DNS, CDNs and layer-4 versus layer-7 load balancers. Here's my summary of the primer's Chinese README.

Scaling SQLite to 4M QPS on one server. This Expensify write-up is the reason I stopped assuming SQLite is only for small apps. Running SQLite behind their own Bedrock layer on self-hosted hardware with 192 physical cores and 1TB of RAM, they pushed parallel reads past four million queries per second, with throughput still climbing. An EC2 x1e.32xlarge topped out around 1.5 million, partly because its 128 vCPUs are 64 physical cores with hyperthreading and the instance seems tuned to drop clock speed under load. The SQLite changes themselves were small; most of the gain came from BIOS and kernel power settings. I go through it in my notes on the SQLite 4M QPS benchmark.

None of these is new, but together they cover most of the questions I hit in the first week of a project: what can I get for free, what data can I use, how do I get things onto a locked-down box, and how far will a simple database actually go. If you have a page that belongs on this list, I'd like to hear about it in the comments.