README
¶
llm-cache
A tiny content-addressed LLM prompt cache. Shows how four stdlib packages compose for a real-world local-state CLI:
| Stdlib package | What it does here |
|---|---|
stdlib/sqlite + stdlib/db |
Schema, INSERT/SELECT/UPDATE/DELETE, db.Transaction |
stdlib/crypto |
SHA256(model|prompt) cache keys |
stdlib/datetime |
NowISO() timestamps, day-based prune cutoff |
stdlib/cli |
Subcommand wiring with put/get/stats/prune |
The get path is the interesting bit — SELECT response and
UPDATE hit_count run inside a db.Transaction so concurrent readers
can't double-count or observe a partial increment.
Run
kukicha run examples/llm-cache/ put --model claude --prompt "hello" --response "hi there"
kukicha run examples/llm-cache/ get --model claude --prompt "hello"
kukicha run examples/llm-cache/ stats
kukicha run examples/llm-cache/ prune --older-than-days 7
The cache lives in llm-cache.db next to where you run the command;
override with --db /path/to/cache.db.
Documentation
¶
There is no documentation for this package.
Source Files
Click to show internal directories.
Click to hide internal directories.