# Mastra Memory
Memory is [[Mastra AI]]'s context-persistence system for agents. It keeps conversations coherent across turns and sessions without overflowing the context window. It combines several mechanisms rather than a single chat-history buffer.
## Mechanisms
- **Message history**: recent conversation messages
- **Observational memory**: compresses conversation into dense, structured observations instead of lossy summaries
- **Working memory**: persistent per-user data carried across sessions
- **Semantic recall**: meaning-based retrieval of relevant past messages (vector search over history)
- **Multi-user threads**: separate memory per thread/user
- Configurable memory processors manage what is retained vs dropped
## References
- https://mastra.ai/docs/memory/overview
## Related
- [[Mastra AI]]
- [[Mastra Agents]]
- [[Mastra RAG]]
- [[Mastra Harness]]
- [[Embeddings]]