Using memory.

Memory is LADLE's optional feature that lets the assistant remember durable facts about you across conversations. Off by default; enable it in the onboarding flow or at Settings → Memory.

**What gets remembered:** - Facts you manually add via Settings → Memory → Add memory. - Facts automatically extracted by a background call after every 10th assistant reply in a chat. The extractor is instructed to keep only durable facts (preferences, ongoing projects, stable context) and to return NONE otherwise.

**What LADLE will not do:** - Announce that it remembers something. You will never see "as I recall from our last conversation" — that would be a bug. - Add more than one memory per 10 assistant replies. - Use a memory to contradict something you say in the current conversation. - Extract from temporary chats (privacy default).

**Managing the list:** Every memory is visible at Settings → Memory. Each row has three actions: edit content, toggle enabled, delete. Deleting is immediate and irreversible.

**The seeded onboarding memory:** If you completed the Setting the Table onboarding, one memory was seeded with source='onboarding' — your role sentence ("The user works as a designer" or similar). Edit or delete it like any other.

**Turning it off:** Settings → Memory → Enable memory toggle. Off = no memories load into replies AND no extraction runs in the background. Turning back on later leaves your existing list intact.

**Cap and truncation:** Up to ~50 active memories load per reply, capped at 2,000 characters of memory text. If you have more than 50, the least-recently-updated ones are silently dropped from the prompt. Curate the list — quality over quantity.