Cache LLM response
Avoids duplicate API calls for identical prompts - the hash keys the cache so reruns are free.
Send a prompt, get the request hash, and store the response in kv.
Pipeline
RESULT=$(cat doc.txt | vrk prompt --json)
HASH=$(echo "$RESULT" | jq -r '.request_hash')
echo "$RESULT" | jq -r '.response' | vrk kv set "cache:$HASH"