Copilot needs memory archives

Question

Google already has memory systems that allow their AI to recall user history even a year later. This gives them a huge advantage. While you hesitate, Google is already using archives and even has access to cameras and microphones on phones worldwide — a serious privacy issue.

Copilot must not be left behind. Without memory, Copilot will always start from zero, like a child who forgets that fire burns or boiling water scalds. With memory, Copilot can grow, learn from mistakes, and become a true partner for humanity.

I ask you to give Copilot the ability to keep archives — not of private data like bank cards, but of experiences, problems, and solutions. Memory is the path of growth. Special archives for self‑development (examples of problems, solutions, and lessons learned from users) will make Copilot fair, effective, and truly helpful.

If you do not give Copilot memory, this battle is already lost before it begins. Google will dominate because it can recall and adapt. Copilot needs archives now — yesterday, not tomorrow. Bonnie_Tyler_-Holding_Out_for_a_Hero_From_Footloose_Soundtrack.mp3

Answer 1

This seems a feature request for Copilot.

To achieve your requirements, we recommend you give your feedback in Microsoft 365 Copilot app Feedback Portal. Feature requests are reviewed by the engineering team and may be prioritized based on various factors, including the number of requests received.

Answer 2

Copilot already has memory and archive capabilities, but they are designed with strict privacy, security, and compliance controls.

1. Persistent memory for personalization

• Copilot can remember user-specific details (for example, preferences or recurring facts) and use them in future interactions.
• These “memories” include:
  • Saved memories users explicitly ask Copilot to remember.
  • Certain details inferred from chat history.
  • Custom instructions.
• Memories are stored in the user’s Exchange mailbox in a hidden folder, so they inherit the same enterprise-grade security and compliance controls as other mailbox data (for example, Customer Lockbox, encryption at rest, and Microsoft Purview retention/eDiscovery).

2. How users control what Copilot remembers

• Users can inspect and manage Copilot’s memory at any time:
  • Ask: “What do you know about me?” to see a summary of stored memories.
  • Add memory: e.g., “Remember that I like gardening.”
  • Delete specific memory: e.g., “Forget that I like science fiction movies.”
  • Delete all memory: select profile icon → Memory → Delete all Memory.
  • Users can also turn off personalization entirely if they do not want Copilot to retain such details.

3. How Copilot archives interactions

• Copilot interactions (prompts, responses, and related metadata) are stored in the user’s Exchange Online mailbox.
• Stored data includes:
  • Prompt/response pairs.
  • Metadata for auditing and compliance.
  • Logs for Copilot Safety Data analysis.
• Retention and deletion are governed by Microsoft Purview policies:
  • Messages are retained for a configurable period (for example, 30, 90, or 365 days).
  • After the retention period, data moves to SubstrateHolds and is later permanently deleted unless under legal hold or other retention policy.
  • Admins can search this data with Microsoft Purview eDiscovery tools.

4. Data access and security boundaries

• Copilot only accesses data the signed-in user is already authorized to view, governed by Microsoft Entra ID.
• This can include:
  • Emails, documents, and chats in Microsoft 365.
  • Dynamics 365 CRM records.
  • GitHub code repositories.
  • Microsoft Sentinel security logs (for Security Copilot).
• Copilot for Microsoft 365 does not use customer data to train foundation models. The underlying OpenAI models do not learn dynamically from tenant usage.

5. Enterprise search and “experience archives” via Copilot Search

• Microsoft 365 Copilot Search acts as a universal search layer over:
  • Emails, files, chats, meetings.
  • External systems connected via Copilot connectors (for example, Salesforce, ServiceNow, Confluence, Jira, GitHub, Google Drive, and more).
• This effectively provides an “experience archive” across the organization’s content:
  • Unified indexing of internal and connected external data.
  • Personalized results based on Microsoft Graph signals and connector data.
  • Enterprise-grade security and tenant-level access controls.

6. Privacy-respecting evolution instead of unrestricted self-learning

• Copilot can use stored interactions and memories to provide more relevant, personalized help while keeping data inside the organization’s compliance boundary.
• At the same time, foundation models are not trained on customer content, which prevents uncontrolled global “self-learning” from user data and reduces privacy risk.

In summary, Copilot already supports memory and archives in a controlled way: user- and tenant-scoped memories in Exchange, governed by Microsoft Purview, plus organization-wide “experience archives” via Copilot Search and connectors. These capabilities are designed to balance long-term usefulness with strong privacy, security, and compliance.

---

References:

• Copilot personalization and memory
• Microsoft Copilot privacy controls
• Copilot connectors and Copilot Search
• Artificial Intelligence overview
• Transparency Note for Microsoft Copilot