Memory
By default, Lucius carries a small set of distilled facts about you between conversations — a budgeting style, a goal you mentioned, a recurring situation he should already know. Separate from your ledger data, which he always has full access to regardless of memory settings; memory is for things that aren’t a database row anywhere.
How it works
After a conversation, Lucius may extract a handful of new facts. Once the list grows large enough, an automatic pass consolidates near-duplicates and stale entries down to a cleaner set. Both steps use a language model call, billed like any other AI usage — point this task at a cheaper model if you like; see Models and cost control.
Everything stored lives in your own local database. Nothing about memory is sent anywhere except your configured AI provider, exactly like an ordinary chat message.
Reviewing what he knows
Settings → AI → Memory shows how many facts are stored and a Review button for the full list — read it like any “what does this app know about me” page. Remove anything wrong from there.
Turning it off
One switch: Memory extraction & injection. Off, Lucius stops extracting new facts and stops recalling any into a reply. It does not delete what’s stored — facts stay on disk until you remove them yourself. Turning memory off and wanting facts gone are two different requests.
Clearing it
From the Review list, delete one fact or clear everything — whether memory is on or off. No need to re-enable it first.
Related pages
lucius/overview— the full capability mapoperations/security-and-privacy— what data exists, and where