Agent Memory Architecture: Building Cross-Session Memory with MemOS

Описание и применение

Интеграция MemOS позволяет сохранять и обновлять предпочтения пользователя по доставке отчетов, обеспечивая согласованное поведение ассистента в разных сессиях.

Задача

The assistant must honor both lasting user preferences and one-off instructions across multiple conversations, distinguishing between them and updating or deleting preferences as needed.

Результат

By integrating MemOS Cloud APIs, the reporting assistant can persist user delivery preferences across sessions, correctly apply one-off instructions, and update or delete preferences as requested, achieving consistent cross-session behavior.

Рабочий процесс

• 1. User sets a default delivery method via a message (e.g., "Send my weekly reports by email by default.")
• 2. Application sends the message to MemOS Cloud using the Add Message API with user_id, conversation_id, and messages array.
• 3. MemOS extracts the preference memory and stores it in MemCube.
• 4. On a new conversation, application calls Search Memory API with user_id and a query about the task to retrieve the current preference.
• 5. The retrieved preference is passed to the model along with the current request (e.g., "Email this report.").
• 6. The assistant applies the preference for defaults but respects one-off instructions for the current task.
• 7. If the user changes the preference (e.g., "Use Slack for future weekly reports."), the application updates the memory via Update Memory API or submits natural-language feedback.
• 8. For deletions, the application calls Delete Memory API with the relevant memory_id or user_id.
• 9. The application monitors async processing via task-status API if needed.
• 10. The assistant continues to honor the updated or deleted preferences in subsequent conversations.

Доказательства

MemOS provides a memory layer for the application to store that preference, retrieve it in a new conversation, and update it when the user changes their mind.

Источник

Send the user’s messages to the Add Message API. A typical request includes a user_id, a conversation_id, and a messages array containing each message’s role and content.

Источник

When the user starts another conversation, call the Search Memory API with the same user_id and a query about the task.

Источник

If your application knows which record needs editing, use Update Memory with its memory_id.

Источник

Use Delete Memory with memory_ids to remove selected memories, or with user_id to clear a user’s memories within the current project.

Источник