What is caching?

You are on Level 1 of 5 of the fak caching ladder.

Audience. Complete beginners — no AI or systems background assumed. By the end you’ll be able to say what a prompt cache is, why an agent gets slower and more expensive without one, and why fak treats keeping it warm as its job.

Short answer. An AI agent forgets everything between turns, so every time you ask it something it re-sends the whole conversation so far just to ask the next thing. A prompt cache lets the provider say “I’ve already seen that part — no need to process it again from scratch,” which makes each turn faster and cheaper. fak’s job is to keep that cache working for you, automatically.

Why does the agent re-send everything?

The model has no memory of its own. Each turn, your agent bundles up everything said so far — your instructions, its answers, the files it read — and sends it all again, plus your new question. The conversation grows, so each turn re-sends a little more than the last. Without a cache, the provider does the full work of reading all of it, every time.

The coat check

Think of a prompt cache as a coat check that resets its closing time each time you touch your coat.

• The first time, you hang up the whole coat — that’s the provider reading your full conversation once and keeping a copy.
• Every time you come back quickly, pickup is fast and cheap — and touching your coat buys you more time before the rack closes.
• But if you stay away too long, the rack is cleared. Next time you start over and hang everything up again.

So the cache rewards staying active and punishes long pauses. (How long is “too long,” and what fak does about it, is exactly what Level 2 covers.)

What do I type?

Nothing extra. You run your agent through fak the normal way:

fak manage claude

Caching is the provider’s feature, and fak looks after it for you from there. There are knobs (Level 2), but you don’t need any of them to benefit.

How can I tell it helped?

One command, after you’ve used a session for a bit:

fak cachevalue report --dev-sessions

It analyzes your own recent sessions and shows what caching saved. On a brand-new machine it may read zero until you’ve run a few sessions — that’s the ledger still filling, not the coat check failing.

Try it

fak manage claude                      # run your agent through fak, as usual
fak cachevalue report --dev-sessions   # later: see what caching saved in your sessions

See also

• Level 2 — Managed cache in practice: the settings, the defaults, and what “managed cache” in the startup banner means.
• What is fak’s managed cache? — the original plain-English page this ladder grew from.
• The fak/DOS glossary — one-line definitions of the words used here.

---

Full ladder (README) · next → Level 2 — Managed cache in practice