Memory

LongTermMemory is the AgentScope interface for persisting user preferences, facts, and key takeaways across multiple turns and sessions. The agentscope-extensions-* repository ships ready-to-use implementations for the major memory backends:

Extension	Backend	Best for
Mem0	Mem0 Platform / self-hosted	General-purpose semantic memory with multi-tenant isolation and metadata filtering
Bailian	Alibaba Cloud Bailian memory service	Cloud-managed memory with rerank / judge / rewrite features
ReMe	Self-hosted ReMe service	Workspace-level memory with trajectory summarization

All three implement the same io.agentscope.core.memory.LongTermMemory interface and are wired into an Agent the same way:

ReActAgent agent = ReActAgent.builder()
    .name("Assistant")
    .model(model)
    .longTermMemory(memory)                       // any of the three implementations
    .longTermMemoryMode(LongTermMemoryMode.BOTH)  // record AND retrieve
    .build();

Choosing an implementation

• Want a single docker run to start locally → Mem0 or ReMe

• Already on Alibaba Cloud Bailian → Bailian

• Need metadata filtering (slice memory by business dimension) → Mem0

• Care about end-to-end conversation trajectory summarization → ReMe

The implementations only differ in initialization parameters and filter semantics; they are transparent to the Agent itself. See each subpage for details.