only-my-mem0ry — Local MCP Memory Server
only-my-mem0ry is a fully local, zero-config Mem0 memory server that adds "long-term memory" to MCP clients running on macOS. No cloud, no separate LLM, no API keys required. Remembering a conversation's context on the device and pulling it back out in the next conversation — that one thing is this project's goal.
Why I built it
Conversational AI is smart, but once a session ends it cleanly forgets the context you just shared. Entrusting memory to a cloud service costs money, piles your data onto external servers, and locks you in over time. So I chose the opposite: a memory server that runs "only on my device."
How it works
The embedding computation that turns sentences into vectors is handled on-device, and storage goes to a local Chroma vector DB. So it works even with no internet and no API key.
Division of roles
The brain role is handled by the client (LLM); this server is responsible only for "memory." Thanks to that, the server itself is lightweight, and whatever LLM client you use can share the same memory.
MCP compatible
Because it follows MCP (Model Context Protocol), different AI clients look at one shared memory store. Stored memories don't just pile up — the design also covers searching them when needed, updating them with new facts, and reconciling duplicates and conflicts.
Operations
launchd manages the lifecycle automatically, so once installed it stays resident in the background and comes up on its own. There's almost nothing to configure — just leave it on and you get a "remembering AI."
What this project set out to protect
By not leaning on the cloud, it gains three things at once: privacy, offline operation, and freedom from lock-in. That's the value only-my-mem0ry wanted to hold onto to the end.
