ai-weekend-builds: a five-volume project catalogue with one API key per volume

What the repository actually contains

This is a catalogue, not a library. The README lists twenty projects across five volumes, each with its own folder, its own README, and starter code described as runnable immediately. Volumes 1 through 3 use an Anthropic key. Volume 4, dated August 2026, is built with Codex and uses an OpenAI key where a project needs one. The README is explicit that the requirement varies: each volume's guide states what it needs. That split matters more than it first appears. If you have an Anthropic key and no OpenAI key, volume 4 is closed to you until you get one, and the reverse is true for volumes 1 to 3. The projects themselves range from a 1 to 3 hour Excalidraw diagram agent in volume 1 to a full-day Freedom Calculator in volume 4 that the README describes as ten thousand simulated versions of your next year, from your own assumptions. Difficulty labels are given per project, and the README tells newer coders to start with the easy-mode projects.

The local-first claim and where it holds

Volume 3 carries the topic tag local-first, and its README section says the tools are local-first wherever possible, so your data never leaves your machine. The hedge in wherever possible is doing real work. Two of the five volume 3 projects are described as fully local in their one-line summaries: Life Admin Brain, where you drop in a lease and policies and ask them anything locally, and Offline Voice Journal, where you talk for two minutes, get a local transcript, and see weekly patterns with no cloud. The other three are not described that way. Money Map takes a bank CSV and reports where the money went plus forgotten subscriptions. Ambient Life Dashboard turns an old tablet into a refreshing scene of your day. Decision Simulator runs ten thousand simulated futures from a big decision. Whether those three call out to a hosted model is not stated in the material I have. If data residency is why you are here, read those three READMEs before you assume the volume-level claim covers them.

The prompt files are the part that is hard to replace

Most tutorial repositories give you code and stop. This one ships the prompts alongside it, and the README treats that as the point: each folder has the exact prompts to get started. That is a different kind of artifact. Code shows you one working shape of a solution. A prompt set shows you the input that produced it, which is what you edit when you want a different output. The volume 2 projects are the clearest case. Screenshot to Code takes a photo of any UI and returns a React component. Content Repurposer turns one post into five platform-native versions. Auto-Skill Builder watches corrections in a chat and produces an installable skill file. In each of those, the prompt is closer to the product than the surrounding script is. The README does not say how the prompts are licensed or whether you may reuse them in your own tooling, which is a gap worth noting if you plan to build on them rather than just follow along.

Getting a project running

The README gives the shape of the workflow rather than a single install command. Each project folder contains a README.md with step-by-step setup, starter code you can run immediately, and the prompts. The repository root states the runtime requirement as Python or Node.js and terminal comfort, and names the key per volume: an Anthropic key for volumes 1 to 3, an OpenAI key for volume 4 where a project needs one. It also gives the navigation path from the root, for example ./vol-4/01-past-you-searcher for the Past-You Searcher and ./vol-3/01-life-admin-brain for Life Admin Brain. What the root README does not give is a package manifest, a requirements file, an install command, or an environment variable name. Those have to come from the individual project README. That is a deliberate structure, since twenty projects with different dependencies would produce an unusable root document, but it means you cannot evaluate the setup cost of a specific project without opening its folder.

Where the repository is thin

The licence field is unknown. No licence file is referenced in the material I have, and the README does not mention terms of use for the code or the prompts. For a repository you clone and run locally that may be tolerable. For anything you intend to ship, redistribute, or use inside a company with a licence review process, it is a blocker until you find an answer. The other thin area is verification. There is no test suite mentioned, no CI configuration referenced, and no releases retrieved. The last push is dated 2026-08-24, so the repository is active, but activity is not the same as a support commitment. There is no issue tracker policy described, no changelog, and no stated compatibility matrix beyond Python or Node.js. If a volume 1 project breaks against a current model API, the material gives you no signal about whether it will be fixed.

How this differs from a framework or a course

The obvious comparison is a general-purpose agent framework, and the difference is in what you receive. A framework gives you abstractions you compose: a client, a tool interface, a loop, and you supply the application. This repository gives you finished applications and the prompts that drive them, with no shared abstraction across the five volumes. Volume 1 includes a Multi-Agent Research Crew and an Autonomous Coding Agent that reads GitHub issues and writes code to fix them. Volume 2 includes an AI Inbox Triage that connects to an inbox and leaves three drafts waiting. Those are end products, not primitives. The trade-off is direct: you get a working thing faster, and you get nothing reusable when you want to build the sixth project yourself. A framework asks you to learn its model first. This asks you to accept someone else's finished shape and modify it. Neither is better in the abstract, but they suit different weekends.

Maintenance cost and what you are signing up for

Every project here depends on a hosted model API, and every hosted model API changes. Prompts that produce good output against one model version are not guaranteed to produce the same output against the next. The repository's own structure reflects this: volume 4 moved to Codex and an OpenAI key, which means the earlier volumes were not simply carried forward onto a new provider. If you build a project from volume 1 or 2 and keep it running, you own the prompt drift yourself, because there is no upgrade path documented in the root README. The cost is not the code, which is small and readable. The cost is re-testing the prompt when the underlying model shifts. On licensing, the honest position is that the material does not state one, so I cannot tell you what you may do with the code or the prompts. Check the repository for a licence file before you build anything you intend to distribute, and treat the absence as unresolved rather than permissive.

Editorial conclusion

Adopt this if you already have an Anthropic or OpenAI key and want a scoped weekend project with starter code rather than a blank file. Skip it if you need a stated licence before you can use the code, or if you want a maintained library rather than a set of tutorials. Before you start, open one project directory and confirm three things: which key its README names, whether the starter code runs on your Python or Node version, and whether the folder contains a licence file the repository root does not mention.

Official sources

1. Issues
2. kju4q/ai-weekend-builds on GitHub
3. README