memory-manager

To preserve AI research memory, you save the run state, goals, and progress using state management. This maintains continuity across long-running workflows by systematically storing working memory, episodes, and procedures for future retrieval.

Recover research state after compaction by retrieving preserved working memory. The system restores the exact run state, allowing the agent to resume experiments or debug effectively without losing prior context or progress.

Manage knowledge retrieval by organizing memory into defined types like procedures, insights, and episodes. This structure enables the agent to systematically reuse prior knowledge and record outcomes for future reference during research.

Yes, this memory management approach works for failed experiments by retrieving the exact state before the failure. The agent recovers the working memory to resume work or debug more effectively without losing prior context.

Yes, you can export shared memory publications from your research workflow. The system prepares shared memory publications by operating across working, episode, procedure, insight, and persona memory types with defined writeback policies.