Spring AI with Llama #2 — Core Concepts: Tokens, Context Window, Prompt & Temperature (Java)

Why does the same question give you a different answer every time? And why is one reply two sentences and the next six paragraphs? In Chapter 2 of Spring AI with Llama, we go under the hood: tokens, the context window, prompts/messages, and temperature — the four concepts that separate developers who fight the AI from those who control it. Then we put them to work: two new endpoints on the SmartHR Assistant — /hr/ask/precise for short, deterministic policy answers, and /hr/ask/creative for open-ended brainstorming — both tuned with Spring AI's ChatOptions. Everything runs 100% locally with Ollama and Llama 3.2. No OpenAI account, no API keys, no cloud bills. ⏱️ What you'll learn in this chapter: • What a token actually is (and why 128K tokens ≈ a whole novel) • The context window — the shared budget for input + output • How a Prompt is really a list of Messages (System, User, Assistant) • How temperature controls randomness — and which value to use when • Capping response length with maxTokens • Building /ask/precise (temp 0.0) and /ask/creative (temp 0.9) with ChatOptions 🔗 RESOURCES 💻 Source code (all chapters, GitHub): https://github.com/balajich/spring-ai... 📝 Written notes / tutorial for this chapter: https://prompttoapps.com/tutorials/sp... 🧠 Test yourself — Chapter 2 quiz: https://prompttoapps.com/quiz/#spring... ▶️ WATCH THE SERIES IN ORDER Series Introduction:    • Spring AI with Llama — Build Local AI Apps...   Chapter 1 — Hello, Spring AI!:    • Spring AI with Llama #1 — Hello Spring AI:...   Chapter 2 — Core Concepts: (you are here) 🧰 Tech stack: Spring Boot 4.1 • Spring AI 2.0 • Java 25 • Ollama • Llama 3.2 Built and tested on Java 25.0.3, Maven 3.9.16, Ollama 0.31.1.