Agent Memory
Persistent AI-powered memory for agents and applications. Automatically extract, classify, and recall knowledge from conversations.
Agent Memory gives your AI agents and applications the ability to remember. Instead of starting every conversation from zero, Agent Memory automatically extracts facts, events, instructions, and tasks from conversations, stores them in isolated per-user profiles, and retrieves the right context when you need it.
You access Agent Memory through a Worker binding with three operations: ingest conversations to extract memories, recall to search and synthesize answers from stored knowledge, and get a summary to generate a Markdown profile overview for system prompts.
Automatic extraction
Agent Memory uses AI to extract discrete, structured memories from raw conversations. You do not need to decide what to remember — the system identifies facts, events, instructions, and tasks automatically.
Semantic and keyword search
Recall combines full-text keyword search, semantic vector search, and AI-powered synthesis to find and present the most relevant memories for any query.
Memory evolution
Knowledge evolves over time. When a user changes a preference or updates a decision, Agent Memory supersedes the old memory with the new one while preserving the full history.
Profile isolation
Each user or agent gets their own isolated memory profile with dedicated storage, search indexes, and vector indexes. Namespaces provide tenant-level isolation for your application.
Agent Memory is accessed through a Worker binding. Build serverless applications that use persistent memory.
Build AI-powered agents with persistent state, tool use, and real-time communication.
Agent Memory uses Workers AI for extraction, classification, embedding, and synthesis.
Agent Memory uses Vectorize for semantic search across stored memories.