MiniMax Desktop renamed to Mavis, launching multi-Agent team collaboration

According to Beating Investigation and Monitoring, MiniMax has upgraded its desktop Agent product as a whole and rebranded it as Mavis (MiniMax as a Jarvis). The core new capability is Agent Teams: users can create multiple Agents with different roles to form a team and collaborate to complete complex long-form tasks that a single Agent is unable to handle. At the same time, the previously separate API and Agent subscriptions have been merged into one; the CLI, API, and Agents are fully integrated, with shared quotas.

MiniMax also released a lengthy technical article to explain the design concept of Agent Teams. The article itself was generated by an Agent Team—one Agent simulates a user asking questions, while another answers based on internal technical materials. The article points out four core problems with a single Agent handling long tasks: it may unexpectedly stop midway through task execution; overly long context causes output quality to decline; long tasks block the user’s real-time interaction; and role-playing at the prompt level cannot achieve a true division of responsibilities.

To address these issues, Mavis drives collaboration using a code state machine rather than prompt orchestration. The team defines three types of roles: the Owner is responsible for task decomposition and scheduling, the Worker focuses on execution, and the Verifier independently performs acceptance. The Verifier and Worker form an adversarial mechanism; the context of each role is strictly isolated, and they communicate only through structured summaries. In the article, MiniMax also openly acknowledges that multi-Agent collaboration introduces additional handoff and aggregation costs, but for long and high-risk tasks, this structured overhead trades for delivery outcome certainty.