Hello @Mohamed, Rihan
Thanks for reaching out to Microsoft Q&A.
Based on the current Microsoft Foundry Agent Service Memory documentation, the following quotas apply to a memory store:
- Maximum scopes per memory store: 100
Maximum memories per scope: 10,000
Search memories: 1,000 requests per minute
Update memories: 1,000 requests per minute
These values are documented under the Memory service's Limitations and quotas section.
1. Can the 100-scope limit be increased?
The current Microsoft documentation does not document a quota-request process or a higher service tier for increasing the 100-scopes-per-memory-store limit.
Therefore, based strictly on the published documentation, we should treat 100 scopes per memory store as the currently documented service quota. We would not recommend assuming that a standard quota request can increase this limit.
2. Can multiple memory stores be used to support more than 100 users/scopes?
Microsoft does support creating multiple memory stores. The Memory usage documentation recommends creating a dedicated memory store for each agent to establish clear boundaries for memory access and optimization.
However, the current documentation does not state that multiple memory stores can be combined/sharded for the same agent specifically to bypass the 100-scope-per-store quota.
Therefore, we cannot confirm from the documentation that distributing users across multiple memory stores is a supported scaling mechanism for exceeding the 100-scope limit.
3. Can the same agent/application use multiple memory stores?
The documentation confirms that memory can be attached to a prompt agent and that memory stores can be created and managed. It also describes a memory store as the persistent storage boundary for an agent.
However, the current Microsoft documentation does not explicitly document a configuration in which one agent is simultaneously associated with multiple memory stores.
Therefore, I would not recommend stating that this configuration is supported unless it is confirmed by the Foundry Agent Service product team.
4. How is user-level isolation intended to work?
The documented mechanism for separating users within a memory store is the scope parameter.
Microsoft states that the scope parameter segments memory across users to provide isolated experiences. When using the memory search tool, automatic scope resolution can also be performed using scope = {{$userId}}.
Therefore, the documented model is essentially:
Memory store → scopes → memories
with each user's memories isolated through their scope.
5. What should be done if the application requires more than 100 scopes?
Based strictly on the current Microsoft documentation, there is no documented configuration for extending a single memory store beyond 100 scopes, and there is no documented architecture that uses multiple memory stores as a sharding mechanism to overcome this limit.
For an application requiring more than 100 user scopes, I recommend validating the intended architecture with the Foundry Agent Service product team before implementation rather than assuming that multiple memory stores can be used to bypass the quota.
Also, Memory is currently a public preview capability, so its quotas and supported capabilities may evolve.
Please refer this Microsoft documentation:
Memory in Microsoft Foundry Agent Service (preview)
Create and use memory in Foundry Agent Service (preview)
Foundry Agent Service limits, quotas, and regional support
Microsoft currently documents a maximum of 100 scopes per memory store. There is no documented quota-increase mechanism for this limit, and the documentation does not currently confirm using multiple memory stores or attaching multiple stores to the same agent as a supported way to exceed the 100-scope limit.
I Hope this helps. Do let me know if you have any further queries.
If this answers your query, please do click Accept Answer and Yes for was this answer helpful.
Thank you!