mem9-ai/mem9

What it solves

AI agents typically lose their context and memories between sessions or when moving across different machines and platforms. mem9 provides a persistent, shared memory layer that allows agents to recall facts and context across restarts, different devices, and even across different agent runtimes (e.g., Claude Code, Dify, and custom clients).

How it works

mem9 operates as a server-backed memory system. Instead of agents storing data in local notebooks or prompt files, they interact with a mem9 server via a REST API to store, search, and update memories. The system supports hybrid recall—combining semantic (vector) search and keyword search—and can be deployed as a hosted API or self-hosted using backends like TiDB, PostgreSQL, or db9. It also features "Space Chains," which allow for the composition of ordered multi-space recall and routing pipelines.

Who it’s for

It is designed for developers building AI agents and coding assistants who need their agents to maintain long-term memory across sessions and platforms, as well as teams managing a fleet of agents that need to share a common knowledge base.

Highlights

• Cross-Platform Persistence: Memory survives session restarts and is shared across different agent runtimes like OpenClaw, Hermes Agent, and DeepSeek Harness.
• Hybrid Recall: Combines semantic and keyword search for more accurate memory retrieval.
• Stateless Integrations: Keeps agent plugins thin by moving storage, search, and ingest policies to the server.
• Visual Dashboard: Provides a way to inspect and manage stored memories without needing to build custom internal tools.
• Flexible Deployment: Available as a managed hosted API or self-hosted with multiple database backend options.