v0.22.0

SQL Sage 0.22.0 - choose how hard the AI thinks: a reasoning-effort selector next to the model picker (only the levels your model supports) and /effort for a single answer; the model list now comes live from your Claude Code installation; security hardening from an external audit: queries that read from linked servers, files or external sources always ask for confirmation, results you agreed to share with one AI provider are never re-sent to another without asking, Run only executes in the window SQL Sage opened, a more truthful audit trail, chat history encrypted on disk with a retention setting; fixes for chats or probes that could hang when the AI CLI left a helper process running.