BUKU PI

English

Bahasa Indonesia

English

Bahasa Indonesia

Appearance

CORE CURRICULUM · 5 modules · 14 lessons + a working-principles section

Buku Pi main path ​

This section handles a complete, continuous, and already-verified learning process. If you want to practice right away, choose a platform entry point from your first success in 30 minutes; if you want to read systematically, start with the Introduction, the guidelines, and the Prologue. Readers who can already install, log in, and use Pi normally can go straight to real tasks starting from module two.

Personal tweets and notes on usage experience are not turned directly into tutorial conclusions. Content that can enter the main path has to be re-checked for its sources, how it is operated, its risk boundaries, and its success signals. Each module also marks companion exercises and further reading, but neither disturbs the order of the main path. After finishing the basic case studies, you can choose content organization or small code repairs to practice transferring the same method to other tasks.

Opening · Learn first why it is worth reading ​

  1. Introduction: Why read this Buku Pi guide
  2. Ten lasting judgments from 98 tweets
  3. Guidelines and notes for the 2026 open learning edition

The Introduction explains Pi's position, the real obstacles for Mandarin-speaking beginners, and the core principles of this edition; the ten judgments distill the parts of personal learning notes that have held up in later practice into this book; and the guidelines set out the platform path, verification dates, maintenance method, and copyright limits.

Prologue · Meet the person behind Pi ​

Mario Zechner's full career timeline starts with AFX and libGDX in 2009, passes through RoboVM, independent development, and Coding Agent practice, and reaches Pi and Earendil in 2026. This section is not an installation prerequisite, but it can help you understand why Pi stays minimal, open, and extensible.

Module one · Installation and basic setup ​

  1. Pre-installation checks
  2. Installing and running Pi
  3. Login and model setup
  4. Starting from a practice directory

Windows users start here

First complete the Mandarin-language Windows path: installing and running Pi; after passing the verification on that page, go straight on to lesson 3. When you later see “a normal terminal”, keep using Git Bash, and replace the practice directory and the fingerprint commands according to the equivalence table on that page.

Completion marker: You can run Pi in its own practice directory and get one real reply.

Once installation is done, you do not have to stop at once to study maintenance commands; when you need to upgrade the version, switch accounts, uninstall, or handle local data, open Lifecycle management after installation.

Module two · Completing real tasks ​

  1. Your first task
  2. Files and working directory
  3. Sessions and resuming work

Completion marker: You can check the input materials, output files, and task requirements yourself, without letting the Agent grade its own work.

Companion content: CASE 01 · Meeting-notes action list · Licensed translation: The Session you cannot take with you

Module three · Long tasks and context ​

  1. Context and compaction
  2. Introduction to prompt caching

Completion marker: As a task keeps growing, you know how to preserve the key results, and you understand that compaction and caching are two different things.

Companion experiment: CASE 02 · Before and after compaction

Officially licensed translations: How Compaction Works in Pi · Prompt caching in Agents

Module four · Extending Pi yourself ​

  1. Skill, Extension, and Pi Package
  2. Extension requirements and verification
  3. How Subagents divide work

Completion marker: You can choose an extension method based on real needs, and check whether a new capability actually works.

Companion exercises: CASE 03 · Turning a method into a Skill · CASE 04 · Loading a minimal Extension · CASE 05 · Two independent review tracks

Unified principles section · Threading all the terms into one complete running process ​

How Pi works: from a single Prompt to a complete Agent Loop returns Session, Context, System Prompt, Tool, Skill, model invocation, and Compaction to the same chain, then continues the meeting-notes task to complete one real tool cycle. It is best read after finishing modules two, three, and four; if the earlier terms are still scattered, you can also use this big picture to set your direction and then go back to practice the related lessons.

Module five · Building a stable workflow ​

  1. Long-running tasks and VPS
  2. Permissions, isolation, and verification

Completion marker: You build checkpoints, recovery paths, and permission boundaries, so long tasks can be resumed or stopped safely.

Companion exercises: CASE 06 · Recovery from a checkpoint · CASE 07 · Safety review before a task · CASE 08 · Graduation project

When you need to look something up instead of continuing the lessons ​

Open the Reference Guide and browse by topic: the FAQ answers common questions, the Troubleshooting guide starts from symptoms you are already seeing, and the Glossary of popular terms explains concepts you have not met yet. If you want to compare Pi with its forks, read the Comparison of the OMP and Selesai Code paths. Personal experience, original tweet text, and changes in understanding are kept separately in Learning notes and will not be mixed into the lesson conclusions.