You already use a notes app, a calendar, a to-do list, and a chat window with an AI in it. They are four silos, none of them really yours, and the smart one has no idea what is in the other three. The alternative is one self-hosted personal assistant that holds your notes, schedule, tasks, documents, and feeds, with an AI on top that can read and act across all of it. Its heart is a pluggable LLM backend, so you choose the AI. Run it on the Claude API for the strongest quality with zero setup, on Ollama Cloud for a large open model without owning a GPU, or on a local Ollama model for total privacy and zero per-token cost. Switching between all three takes one environment variable. The code for all of that is free. It is MIT licensed and public at github.com/rogjenkins/ryoaa: every module, the assistant loop, and 61 tests that pass offline with no API key. Clone it, run make test, and see whether you like how it is built before you spend anything here. This is the book that explains it. Working code tells you what it does. It does not tell you why the backend is a protocol instead of an if statement, why the assistant loop uses a text tool-calling protocol instead of a vendor schema, or what breaks when you change either. That is what you are buying: the reasoning behind every file, and a workbook that walks you through making it yours. Part 1: the book (12 chapters and 3 appendices). A clear, build-along guide: The architecture of a personal information platform. The core chapter: a pluggable LLM backend, hosted or cloud or local, explained from first principles. Storage and retrieval-augmented generation (RAG) with nothing but SQLite. A module per capability: memory and semantic recall, notes, calendar and reminders, todos and recurring chores, news and weather feeds. A tool-calling assistant loop that works identically on any backend. Customizing and extending it, running it day to day, prompt recipes, troubleshooting, and a glossary. Part 2: the customization workbook. Hands-on exercises that take you from a config tweak to a new module of your own, wired into the assistant and covered by its own tests. The workbook is the part that turns reading into a system you actually run, and it is not in the public repository. You also get the code as a versioned download, so the exact snapshot the book describes stays with you even as the repository moves on. What you get that the free repository does not Why, not just what. Twelve chapters on the design decisions: the backend protocol, SQLite as a vector store, the tool-calling loop, and the trade-offs each one buys you. The workbook. Guided exercises, with the reasoning for each one. The build order. The repository is the finished thing. The book builds it in the order that makes it comprehensible, one capability at a time. Prompt recipes, troubleshooting, and a glossary, for when a local model misbehaves and you need to know whether it is your prompt or your parser. A stable snapshot in PDF and EPUB, readable away from a terminal. Who it's for. Developers comfortable in a terminal: you can clone a repo, make a virtualenv, edit a config file, and run make. You do not need to be an AI researcher. You should want to understand and customize your tools, not just install them. Not for you if you want a one-click app with no code, or you are not willing to run anything locally. What makes it different You own the code and the data. Plain SQLite files on your disk. Back it up by copying a folder. You choose the AI. Claude, Ollama Cloud, or a model on your own machine, swapped with one variable. It is a real design, scaled down. The architecture is a single-machine distillation of a production personal-inference platform, not a toy. Built to be changed. Every module is one file, and the final third of the book and the workbook are about making it yours. FAQ The code is free. Why would I pay for this? Because the code is the easy half. Anyone can read 2,900 lines of Python; far fewer can tell you why it is shaped that way, what the alternatives cost, and which parts you can safely tear out. The book is twelve chapters of that reasoning, plus a workbook of exercises that is not in the repository. If you clone the repo and never buy anything, that is a fine outcome and the code stays MIT forever. Do I need an API key? Only for the hosted paths: Claude, or Ollama Cloud. The local path needs no key at all, and there is a built-in demo mode that runs with no key, no Ollama, and no network, so you can try the whole system before choosing anything. Do I need a GPU? No. The hosted paths need no hardware at all, and Ollama Cloud exists so you can run a large open model without owning one. Running locally, a capable CPU handles small models and a GPU helps with larger ones. What language is the code? Python 3.12 or newer. Minimal dependencies (httpx, FastAPI, uvicorn, pydantic), all MIT, BSD, or Apache licensed. Is it really self-hosted? Yes. On the local backend nothing leaves your machine except keyless news and weather requests. Will it stay updated? The design is provider-agnostic on purpose. Swapping in a new model is a config change, so it does not go stale when models change. The book also shows you where to look up current model names, because those expire. Try it backwards if you like. Clone github.com/rogjenkins/ryoaa, run make setup and then make test, and have the assistant answering on your own machine this afternoon. If you get there and want to know why every piece is the shape it is, come back and buy the book. Backed by Gumroad's 30-day money-back guarantee, which on a book with a free codebase attached is close to a formality: you can evaluate the entire system before paying anything.