bx-ai-memory

Add persistent memory to BoxLang AI agents using the aiMemory() function. It supports typed memory abstractions like windowed, summary, and session, with persistence options including file, cache, JDBC databases, or vector stores.

Isolate AI conversation memory in multi-tenant agents by passing userId and conversationId parameters to the aiMemory() function. This tenant-aware scoping keeps per-user and per-conversation data segregated across shared pipelines.

You can use vector memory for semantic search in BoxLang RAG scenarios. The aiMemory() function supports vector stores like Chroma, Pinecone, Weaviate, and in-memory vector options for retrieval and persistence.

BoxLang supports multiple memory types for managing AI conversation context: windowed, summary, session, file, cache, JDBC, and vector. Each type offers distinct persistence and retrieval behaviors for different agent requirements.

Configure a windowed memory in BoxLang by calling aiMemory() with the windowed type and setting a maxMessages limit. Attach the configured memory object to your agent to maintain a rolling context window.

Use database persistence, like JDBC or vector stores, for durable cross-session AI agent context, and in-memory storage for temporary, fast-access data. The aiMemory() config parameter lets you select the appropriate persistence option per memory type.