I built Mnemory, an open-source memory layer for long-running AI agents.
The goal is to make agent memory more structured than “put everything into a vector DB”. Mnemory stores facts, preferences, episodic/context memory, TTLs, importance, user/agent scoping, and artifact-backed long-form memory. It also exposes an MCP server interface.
I built it because I kept running into the same problem with agents: durable facts and short-lived context need different treatment, but many systems collapse everything into one retrieval bucket.
Also I wanted to have one memory system that I can just plug and play into any system.. Opencode, OpenWebUI, Openclaw, Hermes, Cognis, etc.
I’d be interested in feedback from people building personal agents, long-running automations, or memory systems for LLM apps.
You can setup claude-code integration (https://github.com/fpytloun/mnemory/tree/main/integrations/c...) and configure hooks for remember/recall. This is transparent for LLM and does not need anything else.
Anyway it is still useful to configure mnemory MCP too, that will provide LLM tools to work with memories directly and ships with instructions out of the box. You can use `INSTRUCTION_MODE` to customize (passive, proactive/default, personality).
I built Mnemory, an open-source memory layer for long-running AI agents.
The goal is to make agent memory more structured than “put everything into a vector DB”. Mnemory stores facts, preferences, episodic/context memory, TTLs, importance, user/agent scoping, and artifact-backed long-form memory. It also exposes an MCP server interface.
I built it because I kept running into the same problem with agents: durable facts and short-lived context need different treatment, but many systems collapse everything into one retrieval bucket.
Also I wanted to have one memory system that I can just plug and play into any system.. Opencode, OpenWebUI, Openclaw, Hermes, Cognis, etc.
I’d be interested in feedback from people building personal agents, long-running automations, or memory systems for LLM apps.
I’ve come across your app before. What I’m confused is I still have to instruct Claude to save to your app isn’t it?
You can setup claude-code integration (https://github.com/fpytloun/mnemory/tree/main/integrations/c...) and configure hooks for remember/recall. This is transparent for LLM and does not need anything else. Anyway it is still useful to configure mnemory MCP too, that will provide LLM tools to work with memories directly and ships with instructions out of the box. You can use `INSTRUCTION_MODE` to customize (passive, proactive/default, personality).