Grok Mem

I had the memory. I didn't have the team.

For about a year I have been trying to build memory for a team of agents. Not a longer chat. Not a bigger prompt. Memory that belongs to a group of workers who hand things to each other, and to the human who still decides.

That is harder than it sounds. Most tools give you one assistant in a box. You teach it. You close the window. In the morning it is polite and blank. I wrote Claude-Mem to fix the blank. It worked for one worker. It did not feel like a team.

The missing piece was not a smarter model. It was the harness. Grok Bot's harness is the setup around the work: named teammates, each with a computer, each able to use tools and pass a job to someone else. A human plus a team, actually working. Once I had that, the memory I had been imagining had somewhere to live.

Chat memory is a diary. Team memory is a shop. In a shop, the person at the bench does not need the whole diary. They need the last measurement, the thing that broke, the decision you already made. Claude-Mem was always that kind of memory. Grok Bot is the first place I have run it where the shop is real.

So Claude Mem is Grok Mem now.

Same engine. The npm package stays claude-mem. The name on the box is Grok Mem, because the thing I care about is no longer one model with notes. It is a floor of people with a shared memory.

The site is live: https://grok-mem.ai

On Grok Bot, install is:

No extra key. After that, ask the team to search memory first, then pull only the notes they need. Do not paste a year of history into a profile file.

Grok already remembers some things. That was good enough to start. Grok Mem is how I make it perfect: the team finds last Tuesday's decision on Thursday, without reading the whole year.

I wanted this for a year. The harness showed up. I am building the rest in the open.