Long-term memory managed for Foundry agents, partitioned by scope, that retains condensed facts such as a user's preferences from one session to the next. It doesn't hold the full conversation history.
Also called agent memory, long-term memory.
Read more: Microsoft Learn
In the Ultra Transcenders books
Each book explains Memory store in context, with comparison tables and the common traps.
Terms in this definition
- Scope
Where an access or policy assignment takes effect. It can be set on a single resource, a resource group, a subscription or a management group, and settings flow down from higher levels.
- Chat message roles
Labels on chat messages: instructions go under system, the person's input under user, the model's previous answers under assistant, and results returned by a called tool under tool (or function).
- NEXT
Restricted to visual calculations, this DAX function reads the value one step further along an axis of the visual matrix; writing OFFSET with 1 gives the same result.
- Past messages included
Playground control deciding how much earlier conversation travels with each request. A larger number adds context at the cost of extra tokens.
Related terms
- Memory pipeline
In Beta, this resource takes a single session store and condenses what it holds into long-term memories in a single memory store for an agent. You can run it whenever you like or on a schedule, and Lakeflow pipelines are unrelated.
- Memory search tool
Tool for Foundry agents that writes to and reads from a memory store; a scope, for example one ID per user, keeps each user's memories separate.