Skip to main content
AI memory gives the local assistant lasting context, so drafts sound like your team and respect your conventions. There are two scopes: project memory for one engagement, and global memory (house style) that applies everywhere.
AI memory: project memory and house style

AI memory: project memory and house style

​
Project memory

Notes the assistant should always read for this project: scoping context, client preferences, wording to avoid. Injected into every prompt for that project.

​
House style

Two things shape your team’s voice:
  • Global memory entries you pin as house style (“team notes to respect”), injected into every draft.
  • Your team’s own reviewed issues. The assistant learns from issues you have verified, readied, reported or fixed (never from unreviewed AI drafts), and matches their tone, depth and structure on new findings. IP addresses are stripped from these examples so one target’s data never leaks into another’s draft.

​
What the assistant can and cannot do

It drafts from evidence and your style, under your review. It will not run or suggest attacks, mark anything verified, or publish without your approval, and it only ever uses the local model. See Local AI for the guarantees.
The fastest way to teach house style is simply to review and verify a few issues the way you want them written. The assistant picks up your structure from those, no configuration needed.