Walrus Memory
#12 in Dokumenten- & Wissensmanagementwalrus · siet 2026-06-03 · 2× · tolest 29. Juni 2026
16
Momentum
Walrus Memory is a portable, decentralized memory layer for AI agents, officially launched in June 2026 by the Walrus Foundation (Mysten Labs). The platform decouples agent context from individual sessions or providers: memories are encrypted and stored on the Walrus Verifiable Data Platform network, ownership is managed via Sui smart contracts, and memories can be retrieved across sessions and apps. Developers integrate the system via Python and TypeScript SDKs with two API calls (remember / recall); native MCP support and first-party plugins for OpenClaw and NemoClaw are included. The product is currently in beta and actively being developed.
Momentum-Verloop
04.04.03.07.
Features
| Agent Integration (Native Support) | Native MCP support; first-party plugins for OpenClaw and NemoClaw; integration with Vercel AI SDK; out-of-the-box connectors for Claude, ChatGPT, and Gemini; SDKs for Python and TypeScript |
| Input Formats (PDF, Office, Video, etc.) | Raw data as binary blobs (any data type); explicitly mentioned: text logs, structured datasets (Parquet, CSV); general unstructured data of any kind |
| AI Models (Claude, GPT, Proprietary) | Claude, ChatGPT (OpenAI), Gemini (Google), and other leading AI platforms are natively supported; model-agnostic thanks to portability across providers |