Scoping
- Screen context is global. A live session follows the anchored display — the screen your Lace window is on. Both chat and review surfaces can read the live context.
- Drafts are preserved. Draft input is keyed by chat.
- Artifacts belong to their chat. Published prototypes are linked back to the chat that created them.
Session lifecycle
- Start session in the app header, or press ⌘S while Lace is focused.
- Auto-start on review open in Settings starts a session when you enter a review (off by default). Review surfaces join an already-live session instead of starting a second one.
- Stop from the header, ⌘S, or by leaving the review surface. An explicit stop does not block auto-start while screen context is enabled — turn the preference off to stay stopped.
- Sessions also close after 5 minutes of idle.
- Only one session is live at a time. Starting a session elsewhere stops the previous one; recorded data is kept.
Session flow
While a session is live, Lace records the ordered screens and interactions into a session flow. That record is what chat and MCP use when you ask what happened earlier — not a one-off screenshot cache.- Agents find sessions with
search_sessions, then load a flow withget_session_flow(skeleton first; elements and screenshots only when needed). - Sessions are private to you by default. You can share a session with your organization when you want teammates’ agents to see it (same idea as opting a discussion into MCP).
Memory
Lace keeps memory separate from sessions.- User memory stores durable preferences for the current user and organization, such as communication style or recurring constraints.
- Project memory stores durable project context, including current goals, ongoing tasks, and constraints to respect.
- Chat memory stores recent conversation messages for the current chat. Older messages may be compacted into summaries so Lace can still reference earlier context.