Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
[This article is prerelease documentation and is subject to change.]
Memory in the new experience in Microsoft Copilot Studio lets an agent remember details from interactions and apply that context on future ones, so it can deliver more relevant and consistent results over time. An agent with Memory turned on maintains a separate memory store for each user. Use Memory to give your agent persistent knowledge that improves every interaction it has.
Note
This article reflects the new experiences for creating agents and workflows in Microsoft Copilot Studio. The new agent experience is currently available as a production-ready preview. Learn about the two agent experiences in Classic vs. new agent experience. The new workflows experience is in public preview. Learn about the two workflows experiences in Workflows overview.
- Public preview features aren't meant for production use and might have restricted functionality.
- Production-ready and public previews are subject to supplemental terms of use.
- Some capabilities available in the classic experience aren't yet available in the new experience.
- Agents and workflows created in the new experience can't be converted to the classic experience.
With Memory turned on, agents can:
- Recall relevant details from previous interactions
- Apply learned patterns or corrections to future interactions
- Provide more contextual, personalized, and efficient responses
Memory is a controlled, agent-level capability, allowing makers to decide whether an agent uses memory or not.
How memory works
Memory is stored per user, per agent. Each user has their own memory folder for each memory-enabled agent they interact with. The agent reads from and writes to this folder during its work.
The lifecycle is the same for every memory-enabled agent:
- Capture: The agent records signals such as user preferences, observed patterns, or notes about its own work.
- Store: These signals are saved as files in the user's per-agent memory folder, in a tenant-scoped store.
- Apply: On subsequent interactions, the agent reads from this memory to inform its responses or decisions.
Turn Memory on or off
You control whether an agent uses Memory.
To turn Memory on or off:
- Create or open the agent in Copilot Studio.
- Select the Build tab.
- In the components panel, use the Memory toggle to turn the feature on or off.
Important
When Memory is turned on, the agent can remember context and patterns from its interactions to personalize responses and improve efficiency. An agent doesn't share these memories with other users. If a user has no activity with the agent for 28 days, their memories for that agent are deleted.
Turning Memory off doesn't delete stored memories—it only prevents the agent from using them.
Manage your memories
The memory portal shows stored memories for each memory-enabled agent and opens in a new browser tab. You can access the memory portal in the following ways:
- Build tab: From the agent's Build tab, the Memory component links to the management portal. This is the primary path for autonomous agents.
- Preview tab: The first-run response from an agent with Memory turned on includes a link to the management portal.
In the portal, for each memory-enabled agent, users can:
- View memories: View all stored memories for the agent.
- Delete all memories: To delete all memory entries for the agent, select the three dots (…), and then select Delete all memories. Confirm the deletion when prompted.
For conversational agents, users can also manage memories in chat using natural language. For example, asking the agent to describe what it remembers, update a specific memory, or forget something.
Tip
To delete or update a specific memory, ask the agent in chat. To clear all memories at once, use the portal.