Custom instructions vs memory.

LADLE has two persistent-config features that new users often conflate: Custom instructions (you write) and Memory (LADLE observes). They both affect how the model responds; they operate differently.

Custom instructions

Settings → Behavior → Custom instructions. What you write here gets added to every chat's system prompt, in every project. It's your persistent voice for the model.

Good examples: - "Respond in a dry, matter-of-fact tone. Skip pleasantries." - "Default to short bulleted answers. Elaborate only when asked." - "I'm a physician — assume medical terminology in questions is intentional." - "Never open with 'Great question!' — jump straight to the answer."

Custom instructions are what YOU write. Manual. Edit anytime. Applies to every chat.

Memory

Settings → Memory. LADLE observes durable facts from your conversations and stores them here. Extraction fires after the 3rd, 13th, 23rd, … assistant reply in a chat — a background Haiku call looks at the exchange and pulls out one durable fact (or NONE if nothing qualifies).

Examples of memories that get extracted: - "Working on a mobile app for cyclists" (mentioned in a project-planning chat). - "Prefers British English" (came up in an editing chat). - "Runs a small nonprofit focused on adult literacy" (from an intro chat).

Memories are what LADLE OBSERVES. Automatic. You can edit, delete, or add manually from Settings → Memory. Applies to every chat unless the memory toggle is off.

Where they overlap

Both feed the system prompt. Both are user-editable. Both persist across chats. Both surface in the "who is this user" context the model uses when generating a reply.

Where they differ

- **Write path.** Custom instructions = you write. Memory = extracted from chats. - **Content style.** Custom instructions tend to be preferences/rules ("respond this way"). Memory tends to be facts ("this user is a physician"). - **Toggle.** Memory can be turned off entirely (no future extractions, existing memories don't apply). Custom instructions are always applied when set.

When to use which

- **You know exactly what you want the model to do differently.** Custom instructions. Write it once, benefit forever. - **You want the model to remember things about you without you having to spell them out.** Memory. Turn it on, forget about it, it'll gradually accumulate context. - **You want both.** Common. Custom instructions for style rules; memory for facts about your life/work.

Turning memory off — why some users do

Some users prefer explicit control over what the model knows about them. Turning memory off keeps the model's context stable — it only knows what's in the current chat + custom instructions + your latest message. Zero surprises about what it remembers.

Either is valid. Both are user-preference; both are visible.