PTENES
Skip to content
INEMA COURSE · PROMPT CACHING

🧠 cccache — Prompt Caching in Claude Code

Really understand the cache running behind the scenes when you chat with Claude Code — what gets reused, what breaks, and how to pay 10% instead of 100% from the context.

Stable prefix (tools → system → history) tools system history writes 1× CACHE processed prefix turn 1 · reads turn 2 · reads turn 3 · reads turn N · reads each read ≈ 0.1× the input price
Start with Track 1 →
0%
0 of 0

What you'll understand here

The Claude API is stateless: with every message, the entire conversation (tools + system prompt + history) is sent again. The prompt caching keeps that already-processed beginning and reuses it—instead of reprocessing everything from scratch every turn. It’s a prefix matching: the cache key is the exact bytes up to a cutoff point; a single different byte in the middle invalidates everything after it.

The terms, made simple

Hit, miss, write, TTL, breakpoint, cache_read_input_tokens — each one defined at its first appearance.

What breaks the cache

Change the list of tools, dates in the system prompt, nondeterministic order — the silent invalidators and how to find them.

The real savings

Read ~0.1× · write ~1.25× · uncached 1×. When the cache pays off, and when compacting is more worthwhile.

The 3 tracks

How to use this course

  • Mark each topic as read — the module, track, and course progress bars all advance together.
  • Highlight and annotate in the text itself — select a segment to highlight, annotate, or mark as a question.
  • Open “My journey” — see progress, questions, and notes, and pick up where you left off.
  • Adjust the appearance — theme (dark/light/sepia/focus/high contrast), font, size, width, and line spacing.
  • Export your progress — download a .json with your progress and import it on another device.
  • 100% offline — opens directly from the file, without a server. Your progress stays only in your browser.
Start now →