liveBook · Manning

Distributed memory

As explained in AI Governance

Memory whose data is stored across multiple services or components rather than in one location.

When an agent sends personal data to several services, each may retain it independently. Distributed memory makes retention and erasure harder: data can remain beyond the main application database, and downstream services may repurpose it. section 5.6.3

Operational safeguards include:

  • Set retention timelines for agent data and use deletion scripts to purge old caches. section 5.6.3
  • Make working memory, tool-call logs, saved outputs, and longer-term memory searchable and purgeable when handling erasure requests. section 5.6.3
  • Apply retention requirements to agents receiving the data, unless it is needed for further purposes. section 5.6.3

Appointment data across services

Booking a doctor’s appointment can send health data to calendar, transit, and insurance services. Each may retain it independently, so erasure must reach every service the agent touched. section 5.6.3

Chat transcript retention

A chatbot that stores transcripts for learning might be configured to delete records older than 60 days. section 5.6.3

How AI Governance covers it

How deep it goes
Traces data across working memory, tool logs, saved outputs, and downstream services, then recommends searchable stores, retention schedules, and deletion scripts.
Examples
Uses a medical appointment flowing into calendar, transit, and insurance services, plus chat transcripts deleted after 60 days.
What you can do after
Configure retention timelines and deletion scripts, and identify the agent and downstream stores an erasure request must reach.
How it connects
Connects persistent agent memory with MCP, A2A, data retention, purpose limitation, and the right to erasure.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of AI Governance. The section links point to the parts of the book it draws on.