Walrus Memory：跨会话保存智能体记忆

A portable memory layer that makes AI agents reliable across apps and sessions. Persistent, verifiable, and fully under your control.

Walrus Memory Portable agent memory — take your agent's memory anywhere. Walrus Memory enables AI agents to operate reliably across apps and sessions, without losing context. Portable, verifiable, and fully controlled by you, it's the memory layer that lets agents handle complex workflows and coordinate using data they can trust. Walrus Memory is currently in beta and actively evolving. While fully usable today, we continue to refine the developer experience and operational guidance. We welcome feedback from early builders as we continue to improve the product. For AI Agents Single-file guide: Read SKILL.md for a complete integration reference (install, configure, API surface, troubleshooting) LLM-friendly docs: llms.txt — structured overview following the llmstxt.org standard Install pnpm add @mysten-incubation/memwal Peer dependencies (install as needed): pnpm add @mysten/sui @mysten/seal @mysten/walrus ai zod Quick Start import { MemWal } from "@mysten-incubation/memwal"; const memwal = MemWal.create({ key: "your-delegate-key-hex", accountId: "your-walrus-memory-account-id", serverUrl: "https://your-relayer-url.com", namespace: "demo", }); const job = await memwal.remember("User prefers dark mode and uses TypeScript."); await memwal.waitForRememberJob(job.job_id); const memories = await memwal.recall({ query: "What are the user's preferences?" }); await memwal.restore("demo"); Documentation Full docs at memory.walrus.xyz Docs source of truth: docs/ Docs site entry points: What is Walrus Memory? Quick Start SDK Quick Start Relayer Overview SDK API Reference Contributing Report security issues by email to [email protected]. Do not report security issues through GitHub or Discord. See SECURITY.md. We want to be explicit about this while Walrus Memory is in beta: feedback, bug reports, docs fixes, examples, and implementation contributions are all welcome. If you spot rough edges or missing guidance, please open an issue or send a PR. Run the Repo Locally Use Node.js 22.23.4 (pinned in .nvmrc) for repository development. With nvm, run from the repository root: nvm install nvm use pnpm install Important: Build the SDK first — apps depend on its compiled output. pnpm build:sdk Then start the surface you need: pnpm dev:app pnpm dev:noter pnpm dev:chatbot pnpm dev:researcher For the full step-by-step setup guide, see: Run the Repo Locally Exports Entry Description @mysten-incubation/memwal Default client (MemWal). The relayer handles embedding, encryption, Walrus upload/download, retrieval, and restore. @mysten-incubation/memwal/manual Manual client flow (MemWalManual). You handle embedding calls and local SEAL operations. The relayer still handles upload relay, registration, search, and restore. @mysten-incubation/memwal/ai Vercel AI SDK integration - wraps MemWal as middleware for use with streamText, generateText, etc. OpenClaw / NemoClaw Plugin @mysten-incubation/oc-memwal — a memory plugin for OpenClaw agents. It gives OpenClaw agents portable, verifiable memory through Walrus Memory with automatic recall and capture hooks. openclaw plugins install @mysten-incubation/oc-memwal Plugin Quick Start How It Works Reference How It Works Scope - Each memory operation runs inside an owner + namespace boundary Store - The relayer embeds, encrypts, uploads to Walrus, and stores vector metadata in PostgreSQL Recall - The relayer searches by owner plus namespace, resolves matching blobs, and returns plaintext results Restore - The relayer can incrementally rebuild missing indexed entries for one namespace License Apache 2.0 — see LICENSE