Using temporary chats.

Toggle the "Temporary" chip in the composer to make a chat temporary. The reply streams normally. Nothing about the request or the reply is written to LADLE's database — no chat row, no messages, no artifacts. Refresh the page and the conversation is gone.

Temporary chats still count against your usage cap — the tokens were processed by Anthropic's API, so they're metered the same as a regular chat. Meal donation still accrues from your subscription regardless of chat type.

The critical distinction: "temporary" means LADLE doesn't store the chat. It does NOT mean the request never happened. Anthropic's API processed the request and applies its standard retention policy (30-day trust and safety window on the API tier we use, no training on API data). If you need "not even Anthropic sees it," temporary chats aren't the right tool — a local open-weights model is.

Use temporary chats for: prompt experiments you don't want cluttering history, sensitive drafts you don't want re-openable on another device, questions you'd delete anyway. Don't use them for: work you'll want to reference later (there's no undo), or as a substitute for privacy in a threat model that includes API-level surveillance.

The kitchen panel still works on temporary chats (Cmd+K on any reply) — model + routing info lives in response headers, not the database.