What a Team Remembers: Shared Artifacts, Private Memory, and Lifecycle
In JiuwenSwarm, team memory answers a coordination question that appears after the first successful handoff: what should survive the round, who is allowed to write it, and which participants may read it later? This chapter focuses on the difference between shared artifacts, member-private memory, and shared team memory, then compares temporary and persistent teams. The practical goal is to choose continuity deliberately without treating memory as a complete or neutral copy of everything the team has seen.
Three kinds of continuity
A team can share work without sharing every internal state. JiuwenSwarm describes a shared team workspace for artifacts such as research outputs, analysis data, drafts, and task status. It also gives each member a private workspace for configuration, personal memory, and todos. Team messages and task records connect these areas by carrying assignments, progress, and results.
Memory is a third layer. Personal memory belongs to the individual member. Shared team memory is readable by members but is written by an extractor that the Leader launches at the end of a round. This makes team memory different from a shared folder and different from a larger copy of each member's memory. An artifact is an object the team may use in its current work. Personal memory is a member's own accumulated context. Team memory is a selected, interpreted record of what the collaboration should carry forward.
The access rule is as important as the content. Members can read shared team memory, but they do not all write it directly. The Leader's extractor reads the round's task records and messages, decides what is worth preserving, and updates the shared record. That arrangement gives the team a common reference point while limiting uncontrolled edits. It also introduces interpretation: what survives is the extractor's selection and synthesis, not an exhaustive transcript.
This division protects useful boundaries. A member can maintain personal knowledge without making every private note part of the team's public context. The team can record a decision or lesson without granting every member authority to rewrite it. A report or code artifact can be shared for a task without implying that the agent's private configuration or entire memory is exposed. In other words, collaboration requires a shared surface, not total transparency.
Temporary and persistent teams have different contracts
JiuwenSwarm distinguishes temporary and persistent team lifecycles. A temporary team is suited to a one-off task. Its members have read-only access to the parent agent's workspace memory, and the team has no accumulating team memory. When the team is discarded, it does not become a new source of long-term knowledge. The read-only relationship also protects the parent memory from being polluted by a disposable collaboration.
A persistent team is intended for longer-running collaboration. Members have independent personal memory with read/write support, and the Leader's extractor accumulates shared team memory across rounds. Predefined members can continue their existing workspaces, while newly created members begin with their own team-scoped state.