memtensor/memos provides a memory operating system that adds persistent, multi-modal long-term memory to LLMs and AI agents. It offers a unified API for managing memory as an inspectable graph and requires external infrastructure for self-hosting.
Project overview
It addresses the lack of persistent context in AI agents by unifying storage, retrieval, and multi-modal memory management into composable knowledge bases.
Project type
AI Agent · RAG
Use cases
Knowledge Q&A · Automation
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers and AI engineers building automated agent applications who need to implement cross-task memory and multi-modal retrieval.
Key capabilities
Provides a single API to add, retrieve, edit, and delete memory structured as an inspectable graph.
Natively supports text, images, tool traces, and personas retrieved and reasoned together.
Manage multiple knowledge bases as composable memory cubes for isolation and controlled sharing.
Refine memory with natural-language feedback—correcting, supplementing, or replacing memories.
Powers self-evolving memory across L1 traces, L2 policies, L3 world models, and crystallized Skills locally.
Limitations and risks
Self-hosting the REST service requires setting up and configuring Neo4j and Qdrant.
Self-hosting requires an external LLM provider and embedder configured via API keys.
Getting started
Clone the repository, fill in API keys in the .env file, run docker compose up, and open http://localhost:8000. Setup is rated medium difficulty due to the configuration of external databases.
Alternatives and comparisons
A framework for chaining interoperable components, models, and vector stores to build AI applications, which can be used to construct custom memory and retrieval pipelines.
An open-source platform that provides visual tools for building agentic AI workflows and RAG pipelines with integrated model management and backend-as-a-service APIs.
A collection of tested, open-source AI agent and RAG application examples, including agents with memory and multi-step reasoning, designed for quick-start implementation.
GitHub project description: Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings
README: **MemOS** is a Memory Operating System for LLMs and AI agents that unifies **store / retrieve / manage** for long-term memory, enabling **context-aware and personalized** interact…
README: - **Unified Memory API**: A single API to add, retrieve, edit, and delete memory—structured as a graph, inspectable and editable by design, not a black-box embedding store. - **Mu…
README: MemOS leads across multiple benchmarks — evaluated against mainstream commercial memory products across 5 user memory and 5 agent memory tasks.