{"id":12412,"date":"2026-06-02T12:15:28","date_gmt":"2026-06-02T19:15:28","guid":{"rendered":"https:\/\/devblogs.microsoft.com\/cosmosdb\/?p=12412"},"modified":"2026-06-02T03:01:28","modified_gmt":"2026-06-02T10:01:28","slug":"new-toolkits-for-agent-memories-and-agentic-retrieval-in-azure-cosmos-db","status":"publish","type":"post","link":"https:\/\/devblogs.microsoft.com\/cosmosdb\/new-toolkits-for-agent-memories-and-agentic-retrieval-in-azure-cosmos-db\/","title":{"rendered":"New Toolkits for Agent Memories and Agentic Retrieval in Azure Cosmos DB"},"content":{"rendered":"

Today we\u2019re thrilled to announce the public preview of two new toolkits for Azure Cosmos DB: the Agent Memory Toolkit<\/strong> and the Agentic Retrieval Toolkit<\/strong>. If you\u2019re building AI agents and retrieval-augmented generation (RAG) apps, these toolkits are designed to take you from prototype to production faster, giving your agents durable memory and your RAG pipelines the ability to reason over evidence in multiple passes, all backed by the database you already know and love.<\/p>\n

Both toolkits build on the same foundation: Azure Cosmos DB for NoSQL as a unified store for documents, vectors, and full-text data, with vector search, full-text search, and hybrid search built in. That means no extra vector database to provision, no glue code to keep two systems in sync, and one place to query everything your agent needs. Let\u2019s dig into each one.<\/p>\n

Agent Memory Toolkit<\/h1>\n

What is it?<\/h2>\n

The Agent Memory Toolkit is a Python SDK that helps you add memory to AI agents backed by Azure Cosmos DB. It stores raw conversation history and creates higher-value derived memories, such as summaries, facts, and user profiles, so your agent can recall useful context across messages, threads, and sessions.<\/p>\n

\"Promotional<\/h2>\n

Why is this important?<\/h2>\n

Agents are only as good as what they remember. Without memory, every conversation starts from scratch, your agent forgets what the user told it two messages ago, repeats questions, and can\u2019t personalize. Bolting memory on yourself usually means stitching together a vector store, a document store, and a pile of custom transformation logic.<\/p>\n

The Agent Memory Toolkit gives you that whole stack out of the box. It supports both short-term<\/strong> memory (individual turns during an active conversation) and long-term<\/strong> memory (durable artifacts that are easy to retrieve later), so your agent can hold a coherent thread now and recall stable user context weeks from now.<\/p>\n

Memory types<\/h3>\n\n\n\n\n\n\n\n\n
Memory type<\/strong><\/th>\nDescription<\/strong><\/th>\nCommon use<\/strong><\/th>\n<\/tr>\n<\/thead>\n
turn<\/strong><\/td>\nRaw conversation records for user, agent, tool, or system messages.<\/td>\nReplay recent conversation history or preserve short-term context.<\/td>\n<\/tr>\n
summary<\/strong><\/td>\nA compact summary of one conversation thread.<\/td>\nRecall the main topic, decisions, open issues, and next steps from a long thread.<\/td>\n<\/tr>\n
fact<\/strong><\/td>\nA discrete assertion extracted from a thread, such as a preference, requirement, or confirmed decision.<\/td>\nRetrieve fine-grained knowledge across threads by semantic meaning.<\/td>\n<\/tr>\n
user_summary<\/strong><\/td>\nA cross-thread profile for one user.<\/td>\nPreserve stable user context, preferences, environment details, and constraints across sessions.<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

How does it work?<\/h2>\n

The toolkit supports local development, durable storage, retrieval, semantic search, and memory transformation. Azure Cosmos DB serves as the durable memory store, keeping turns, summaries, facts, and user summaries as JSON documents and supporting query, vector search, full-text search, and hybrid search over all of them.<\/p>\n

Two ways to process memory<\/h3>\n

You can run memory processing in two ways, depending on how much infrastructure you want:<\/p>\n