AI Agents Finally Share One Memory (Claude Mem Explained)

0:00You know, you work on a project in Claude and then you move on to chat GPT and chat GPT doesn't know what Claude did and so you have to start fresh there and maybe the next week you go into Grok Bot or something else and it doesn't know what either of those two did. Well, there's a solution to it. It's called Claude Mem and what it does is it keeps a memory of what all of your agents are doing and allows them to share information. So, you can start a project on one, then move to the other, and then maybe use a third and a fourth and so on

0:25and they all work together. It's called Claude Mem. The creator is here to show you how to use it. Presented by Zapier, the AI automation company. Okay, so on the left here is essentially Claude code and I have just ran my community sweep skill and what that does is it finds all the PRs that are ready to merge, makes sure that they are doing what they need to do, and writes documentation, and follows up with people. I can't figure out what is happening at the end it because Claude tends to write this stuff with no context. So, I'm going to have chat GPT make me beautiful infographics directly

1:07from what Claude did and I'm going to use Claude Mem to do it. So, Claude Mem um get the latest uh from the community sweep we did and make me mobile friendly generated infographics telling me the story of what was done. It came up. Look at how pretty that is. Okay. Community fixes landed. 25 community PRs, 13 contributors, plus two maintainer fixes. At 11:47 we took

1:49stock. We post-hoc dependency broke builds at 12:13 and it's been landing the work so far. So, I'll get details on the next two slides. Okay, but the idea is we obviously we picked a simple illustration of how chat GPT now knows what code what Claude is doing and now Grok bot can also know what both of these are doing and they're all communicating with each other using

2:16Claude Mem, right? Yes, exactly. Exactly. Am I trusting you with all my my agent's memories now by having you put it in your cloud instead of on my computer? Well, the the paid cloud is it's secure. You can try to run your own docker of it as well and like keep it even more isolated. But, security and all that stuff is super important to me. It was the reason I made Claude Mem local in the first place. Um I see. So, I could actually have a version of it that's local on my computer and then all the agents can

2:51access it from my computer. Yes, and there's a server in there, too. Open source that you can install the server and then you can you can set up your own remote stuff. I I am just very much all about the UX of like it just works in the background. That's why Claude Mem was the background observer. I wanted to be workflow agnostic so that you wouldn't have to learn how to do

3:14something new in order to use the tool. Um and that it would just make the way you currently work better. Cuz if you have to learn a tool, then that's a gamble. It's a risk of like, will this help me or will this keep me from having dinner with my family tonight on time,

3:32you know? Can I see how to install it? Yeah. NPX Claude Mem install.

3:53Okay, got it. All right, so now it's asking which agents do I use? So I connect in the ones that I use. Got it. just added T3 code. I just added Deep Sea harness. Um, yeah. Grockbot, I think you said was 3 weeks ago. Um, how do I connect it to Grockbot? Grockbot's right here. Um, and what you do to for Grockbot is you can actually just um, you can go to grockmem.ai,

4:19grock-mem.ai. Um, or you can just install it like this, or you can just tell it install Claude mem. So what are some ways that you use these interactions between the different agents? So very very much been building the work state tracker over here in Claude code. Um, but I also do a lot of work remotely now using the work tab from ChatGPT. So I'll have stuff running here. Um, I'll have

4:50things built. I'll do a lot of research. And I then take that stuff and I use ChatGPT to finalize it. So it's essentially you find the model that's best for each job, you do it in there, and the way they all coordinate with each other is through Claude mem. Right, because I I don't have to do any handoff work anymore because it's taking the handoff work away from me doing it

5:18in the in the notes in the background. My ideas from months ago can come into my new work because I can just say, "Hey, what's up?" And then I don't need to remember that I did it in ChatGPT. I don't need to remember that I did it in Claude or Grockbot. Is the vision for you, is it more than

5:34just you seeing where everything is? Will they also start handing things off? Will Claude hand off a design project over to ChatGPT, and then ChatGPT will bring it back to Claude and say it's ready for you to program? Is that the way you envision it working? What I'm actually imagining is is having like a single orchestrator bot

5:54kind of like how Grok bot works. And that is going to be who I talk to about everything. We need to have a higher level of control in order to make everything smooth. So, I am now focused on being the universal memory adapter for all agents. All right. The way that I discovered this is on the GitHub repo show where every week we look at the top 10 trending repos. If you like this, I want you to see the full top 10 trending and there's a link to it right here. I'll

6:25see you in that one.