Memori Expands Into a Full-Scale Memory Layer for AI Agents Across SQL and MongoDB

"Memori has matured into a full-featured, open-source memory system designed to give AI agents long-term, structured, and queryable memory using standard databases rather than proprietary vector stores. Instead of relying on ad-hoc prompts or ephemeral session state, Memori continuously extracts entities, facts, relationships, and context from interactions and stores them in SQL or MongoDB backends, enabling agents to recall and reuse information across sessions with no manual orchestration."

"The system offers a database-agnostic architecture, enabling developers to use SQLite for local projects, PostgreSQL or MySQL for scalability, or MongoDB for document-oriented needs. Memori automatically detects the backend in use and manages data ingestion, search, and retrieval through specific adapters, all while maintaining a consistent external API. This approach makes it appealing for production workloads where reliability and portability are key."

"Memori's memory engine automatically extracts and categorizes entities into facts, preferences, rules, identities, and relationships. It prioritizes interpretable storage, saving memories in a human-readable format for easy inspection, export, or migration without vendor lock-in. Agents can retrieve information without needing to create SQL queries, as the process is fully abstracted. As Sumanth P explained in response to a community question: Memori handles the storage internally, and the agent can retrieve info through its API without generating SQL directly."

Memori provides AI agents with persistent, structured memory by extracting entities, facts, relationships, and context from interactions and storing them in SQL or MongoDB backends. The architecture is database-agnostic, supporting SQLite for local use, PostgreSQL/MySQL for scale, and MongoDB for document needs, with adapters that auto-detect and manage ingestion, search, and retrieval while exposing a consistent API. The memory engine categorizes memories as facts, preferences, rules, identities, and relationships and stores them in human-readable form for inspection, export, or migration. Retrieval is abstracted through the API, removing the need for manual SQL. LangChain and multiple LLM ecosystems are supported.

Read at InfoQ