The pilot
Prompt caching
A 7-part series on prompt caching, and why it decides what you pay. Everything you need to stop paying full price by accident.
SKILL UNLOCKED · Cache Freeze
Hex's intro, all seven episodes and the summary in one go: 10:53.
Episodes
-
Length 1:19Ep 1: Why your second message is cheaper
6.64 credits, then 0.57: same words, a tenth of the price
-
Length 1:14Ep 2: What the model does with your text
tokens → vectors → attention: the work the cache saves
-
Length 1:16Ep 3: The prefix
same start: ~90% cheaper; change the top and pay again
-
Length 0:51Ep 4: The cache isn't one box
each version gets its own entry
-
Length 1:00Ep 5: What breaks the cache
models, effort, skills, MCP: miss. AGENTS.md: wait for a new session
-
Length 1:17Ep 6: Time to live
5 min by default, 1 h if your tool asks
-
Length 1:06Ep 7: Tools, system, messages
touch the top, rebuild the lot
-
Length 1:29Summary: What have we learned today?
all seven episodes in one recap, and Hex unlocks Cache Freeze
