Skip to content
EN

Agent memory and instructions

Open Settings → Memory to create, edit, delete or move notes between global, project and conversation scopes. The conversation’s Memory panel uses the same editor. The agent’s memory tool uses the same scoped notes; a child does not automatically inherit its parent’s conversation notes.

Saved notes become available immediately. An active conversation receives newer note facts without rewriting its frozen system-memory prefix. A successfully applied context summary advances that memory snapshot.

Open Settings → System prompt to edit instruction sections. Density, style and autonomy controls insert editable text. Global, project and conversation overrides apply in that order; Restore defaults removes only the selected scope’s edits. Saved profile changes apply at a safe request boundary, while already prepared requests stay unchanged.

The preview shows assembled instructions and the captured memory snapshot. When available, the last-request view records prepared system text; it is not proof of provider delivery. File-based instructions and skills retain their own captured resources: use Refresh instructions and skills after changing them.

Bot instructions and memory remain separate in Varin Bot. Ordinary Agent notes and prompt overrides are not injected into Bots.