MAPI

MAPI is a governed memory layer for long-running AI agents.

Most AI systems can retrieve previous information, but they still struggle to maintain reliable continuity across sessions. Decisions become scattered across conversations, summaries and files. When a decision changes, older and newer versions may be mixed together, silently overwritten or returned without context.

MAPI addresses this problem by turning long-term memory into a controlled lifecycle. It helps an agent decide what should be remembered, preserves the source and scope of each memory, and records how important project decisions change over time.

Confirmed knowledge can be restored by a different AI agent in a new session. Updated decisions can replace earlier versions without deleting their history. Conflicts remain visible until they are reviewed, while audit and rollback tools make important changes inspectable and reversible.

The competition demo shows ChatGPT and Codex using the same MAPI memory layer. ChatGPT stores and updates decisions for a fictional booking project, while Codex restores and uses those decisions in separate sessions without receiving the original conversation.

MAPI is an advanced working prototype designed for coding agents, multi-agent systems, private assistants and other workflows where remembering the wrong information may be more costly than asking for human review.