Persist shared state across runs
MemoryStore is a namespaced key-value persistence primitive. Use it for durable shared state; add a retrieval layer when you need semantic long-term memory.
The problem.
In-memory coordination disappears with the process. Some workflows only need durable keys and checkpoints; assistants that recall facts or user context need a separate semantic extraction and retrieval system.
How open-multi-agent does it.
open-multi-agent exposes a MemoryStore interface whose keys are namespaced by agent. Use FileStore or a custom Redis/Postgres adapter for durable state. The same interface underpins checkpoint restore at completed task boundaries. For semantic recall, connect a system such as TencentDB Agent Memory or your own retrieval layer; OMA does not infer, embed, rank, or govern memories for you.
When this fits.
Use MemoryStore directly for durable workflow state, checkpoints, caches, and caller-defined records. Add semantic memory only when the product must extract and retrieve meaning across sessions. If every run is independent, in-run shared context is enough.
Taking this to production?
open-multi-agent is MIT-licensed and free to run yourself. When you need it delivered, integrated, or supported on a deadline, 元定义科技 (YuanASI) offers commercial delivery and support.