MariaDB Agent Memory
Product · Developer tool by MariaDB
Durable state and semantic memory for AI agents built into your database platform.
MariaDB Agent Memory is a specialized data infrastructure solution designed for AI agents to maintain state, facts, and knowledge across sessions. It provides two distinct types of storage: ultra-fast working memory in a distributed cache for short-term task checkpoints and semantic memory within the core database for long-term business knowledge. By storing agent history and embeddings alongside operational data, it eliminates the need for separate vector databases and complex ETL pipelines.
- Languages
- English
- Supported Frameworks
- LangChain, LangGraph
- Latency
- Sub-millisecond
- Storage Engines
- MariaDB Enterprise Server, MariaDB GridGain
Overview
MariaDB Agent Memory is a specialized data infrastructure solution designed for AI agents to maintain state, facts, and knowledge across sessions. It provides two distinct types of storage: ultra-fast working memory in a distributed cache for short-term task checkpoints and semantic memory within the core database for long-term business knowledge. By storing agent history and embeddings alongside operational data, it eliminates the need for separate vector databases and complex ETL pipelines.
Memory Types
- Short-term working memory: Provides a durable, in-memory store for agent state that is shared across sessions and workers.
- Cross-session memory: Stores preferences and account details by namespace and key with server-enforced time-to-live (TTL) automatically.
- Semantic memory: Holds embeddings in a VECTOR column alongside relational rows for single-statement filtering and ranking.
- Framework integrations: Implements checkpointer, store, chat history, and vector store interfaces for seamless backend configuration.
Technical Capabilities
- State persistence: Ensures a task survives a restart and resumes where it stopped, even on a different worker.
- Shared access: Maintains consistency across many users, sessions, and agents reading and writing at once.
- Low-latency caching: Uses MariaDB GridGain for sub-millisecond responses and automated TTL management.
- Metadata filtering: Allows one SQL statement to filter by metadata and rank by meaning without a separate vector store.
Specifications
- Supported Frameworks
- LangChain, LangGraph
- Integration Packages
- langchain-gridgain, langgraph-checkpoint-gridgain, langchain-mariadb
- Latency
- Sub-millisecond
- Storage Engines
- MariaDB Enterprise Server, MariaDB GridGain
Added by AntSpace from public sources (mariadb.com).