TLDR
Walrus and Memory is a decentralized, encrypted memory system that works across multiple AI coding assistants like Claude Code, Codex, and Pi via a single MCP server prompt, solving the problem of siloed memory in individual tools.
Key points
- The creator built a memory system for Claude Code but needed a cross-tool solution for Codex and Pi.
- Walrus and Memory provides a decentralized memory layer that is encrypted and owned by the user.
- Setup requires just one prompt copied from Walrus's homepage to install an MCP server.
- The Walrus MCP server offers tools to remember key information, analyze text for memories, and search memories.
- In a demo, Walrus extracted 20 facts from a text blob and stored them encrypted.
- Memories can be accessed from any tool with the Walrus MCP server, demonstrated with Claude Code and Pi.
- Walrus supports access control with delegate keys for different systems or people.
- It is free to get started and described as fast to set up.
Tools mentioned
- Walrus and Memory
- Claude Code
- Codex
- Pi
- MCP Server
Techniques
- Decentralized memory system via MCP server
- Encrypted memory storage
- Cross-platform memory retrieval
Takeaways
- Walrus provides a portable memory layer that works across multiple AI coding assistants.
- Memories are encrypted and user-owned, not stored on company infrastructure.
- Setup is simple with a single prompt to install the MCP server.
- Access control features allow selective sharing of memories.
Transcript (captions)
Claude Code through my second brain knows everything about me and the projects that I'm working on. The memory system here makes me so productive, but the problem is I'm not just using Claude Code anymore. I'm still using it, but I'm also using Codex and Pi. These tools have their different strengths, so I'm picking the right one for each individual task. But the problem is the memory system that I built into Claude Code in my second brain locks me into Claude Code. So, I needed a solution that allows me to carry all of my memories about myself and my business between the different tools that I'm using. Now, I got to say, I've been looking for a solution to this for quite a while, and recently I came across this tool called Walrus and Memory, and it's perfect for what I'm looking for. It's a memory system that lives outside of any single tool, and your memories are decentralized, so they're encrypted and owned by you, not living on some company's infrastructure. And the best part is it is just one prompt to get started for free. You just copy this from their home page I'll have linked in the description, send this in any AI coding assistant, and it'll set up the memory system. And with their SDK, you can build it into any app as well. So, once you get Walrus installed with that single prompt, you should be able to see the MCP server connected here. And then if we go into the individual tools, we have tools to remember key information, even analyze a larger chunk of text to extract through memories, and then of course search through our memories as well. And so, usually you're going to have to remember things more naturally in a conversation, but for the sake of demo here I'll paste in a bunch of text and I'll just say use Walrus to remember this. And so, it's going to call upon the MCP server to extract key memories from this large blob of text just from some stuff I was working on with my second brain. All right, and then just a little bit later it successfully extracts 20 facts from the text I gave it, encrypted and stored it in my memories. I can now use this across my tools. So, in another Claude Code session for example, I can say search my Walrus memories to tell me about the tech stack for for Dark Factory. This is something specifically I pulled from the text. And there we go. We got the perfect answer. A full text stack and it got this from using the Walrus Memory MCP. And so, we can now go to another coding agent. Like, let's go into Pi, for example. I also have the Walrus Memory MCP server here. And so, I'll paste in the exact same question. No additional setup besides just connecting the MCP server, and I can immediately access all of my memories in Pi as well. And we got the perfect answer yet again. A bit more concise than Claude, but it is the same information that it pulled. And because all your memories in Walrus are encrypted and owned by you, you have a lot of options for access control, creating delegate keys for different systems and even people to access your memories in different ways. And it's just so easy to take this as the memory layer you can now use in any AI tool. So, link in the description, free to get started, and just really fast to get it set up. So, it's really worth a shot.
Jobs for this video
| Stage | Status | Attempts | Last error | Updated |
|---|---|---|---|---|
| summarize | done | 0 | — | 2026-06-24 03:36:04.367114+00:00 |
| summarize | done | 0 | — | 2026-06-24 03:35:45.804641+00:00 |
| transcript | done | 0 | — | 2026-06-24 03:35:22.979932+00:00 |
| transcript | dead | 5 | handler returned RETRY | 2026-06-19 22:15:45.194174+00:00 |
| metadata | done | 0 | — | 2026-06-19 22:00:21.464250+00:00 |