BoxLang AI 3.0 ships with 20+ memory types split into two categories: standard memory (conversation history) and vector memory (semantic embeddings for RAG). Standard types include window, summary, session, file, cache, and JDBC. Vector types integrate with ChromaDB, PostgreSQL pgvector, Pinecone, Qdrant, Weaviate, Milvus, and more. A hybrid type combines recency and semantic retrieval for production use. The framework includes 30+ document loaders for RAG ingestion pipelines, per-call multi-tenant identity routing via userId and conversationId, token management BIFs, and support for multiple memory instances per agent. A complete RAG pipeline walkthrough covers ingestion, querying, and hybrid production setup.
Table of contents
Two Categories of MemoryStandard Memory TypesVector Memory TypesPer-Call Multi-Tenant Identity RoutingDocument LoadersBuilding a Complete RAG PipelineToken ManagementMultiple Memories Per AgentThe aiPopulate() BIF — Structured Memory Without Live CallsWhat's NextSort: