Setting custom instructions.

Custom instructions are your persistent addition to LADLE's system prompt. Whatever you put here gets added to every chat, so you're not re-explaining yourself every conversation. Go to Settings → Behavior.

**What works well as a custom instruction:** - Voice preferences: "Respond in a dry, matter-of-fact tone. Skip the pleasantries." - Format preferences: "Default to short bulleted answers. Elaborate only when I ask." - Domain framing: "I'm a physician; assume medical terminology in questions is intentional." - Things to always/never do: "Never open with 'Great question!' — jump straight to the answer." - Working context: "I write in American English. Round numbers to two sig figs unless I specify."

**What doesn't work well:** - Very long instructions (500+ words). The model attends to them but they eat context and the model may prioritize inconsistently. - Task-specific instructions. Better to put these in a Project (which applies only to a subset of chats) or in the individual chat's opening message. - Instructions that contradict the model's safety training. Won't work; wastes tokens.

**Difference from Memory:** custom instructions are what YOU write, apply to every chat, edited manually. Memory is what LADLE observes about you (facts you mention in conversation) and stores automatically for reference. Both feed the system prompt but are separately editable.

**Difference from Response style:** response style (also in Settings → Behavior) is a dropdown — Concise / Thorough / Exploratory / Precise. It's the pre-baked version of a common voice preference. If none of them fit, custom instructions is where you write your own.