Local model chore: draft a sensitive message in private

The Stack

Tested Against

Prerequisites

• A laptop with ~8GB RAM

Steps

1. 1

   Soften the message on a real local model (offline)

   Install Ollama, pull a Gemma 3 model, and ask it to rewrite a heated message to be calmer. After the download this runs with wifi off, so the contents stay on the machine. CI runs exactly this on gemma3:270m and checks a real draft came back.

   command -v ollama >/dev/null 2>&1 || curl -fsSL https://ollama.com/install.sh | sh
   curl -s http://localhost:11434/api/version >/dev/null 2>&1 || (ollama serve >/tmp/ollama-serve.log 2>&1 &)
   for i in $(seq 1 60); do curl -s http://localhost:11434/api/version >/dev/null 2>&1 && break; sleep 1; done
   ollama pull gemma3:270m >/dev/null 2>&1
   OUT=$(ollama run gemma3:270m "Rewrite this message to sound calmer and more polite, keep it short and sincere: I am extremely frustrated that the invoice is wrong again and I need this fixed today.")
   if [ -n "$OUT" ]; then
     echo "chore3 OK: gemma3:270m drafted a $(printf '%s' "$OUT" | wc -c | tr -d ' ')-char softened message offline (quality fenced)"
   else
     echo "EMPTY OUTPUT"; exit 1
   fi

2. 2

   Use the recommended size and reread it (the quality step, not checked by CI)

   For real messages, pull gemma3:4b or larger for better tone, then always reread before sending. The draft's quality is non-deterministic, so CI never claims it.