MemMachine

Give Your AI Agents Long-Term Memory

MemMachine is an open-source long-term memory layer purpose-built for AI agents and LLM-powered applications. Most AI agents are stateless. They forget every conversation the moment it ends, forcing users to repeat themselves and limiting the depth of personalization your application can deliver. MemMachine solves this with a hybrid memory architecture that combines three complementary memory types: Episodic Memory (graph-based conversational context that persists across sessions), Profile Memory (long-term user facts and preferences stored in SQL), and Working Memory (short-term in-session context). Together, these give your agents the ability to learn, remember, and reason about individual users over time -- turning generic chatbots into genuinely context-aware, personalized AI experiences.

Built for AWS. Integrates with Your Entire AI Stack.

MemMachine deploys natively on AWS and integrates out of the box with Amazon Bedrock and the AWS Strands Agent SDK, so you can add persistent memory to your Bedrock agents without rearchitecting your stack. MemMachine is LLM-agnostic and works with any Amazon Bedrock-supported model. Developers can connect via Python SDK, TypeScript SDK, or RESTful API and have seamless integrations with LangChain, LangGraph, LlamaIndex, CrewAI, n8n, Dify, and FastGPT. Whether you're building a customer-facing assistant, an autonomous workflow, or a multi-agent system, MemMachine fits into your architecture with minimal code changes.

SOTA Memory Accuracy. Production-Ready. Open Source.

MemMachine is released under the Apache 2.0 license and designed for both self-hosted and managed deployments. Run it in Docker, deploy it on your own AWS infrastructure, or use the MemMachine managed cloud platform hosted on AWS. The storage layer is purpose-built for AI memory workloads: Neo4j powers the episodic graph for rich relational context retrieval, while SQL handles structured profile data at scale. MemMachine delivers industry-leading scores on the open-source LoCoMo benchmark, achieving an LLM-score of 0.91 across multi-hop, temporal, single-hop, and open-domain reasoning tasks, while using ~80% fewer tokens than competing memory systems. Real-world use cases include CRM agents that recall full client history, healthcare navigators that track treatment context, financial advisors that remember portfolio preferences, and writing assistants that learn your style guide. With a growing open-source community, comprehensive documentation, and integrations across the most popular AI frameworks, MemMachine is the SOTA, production-grade agent memory layer for AWS.