M memwal
decentralised agent infrastructure

Persistent memory
for AI agents

Walrus-backed LangGraph checkpoint storage on Sui. Store, retrieve, and manage agent memory across sessions — decentralised, verifiable, permanent.

walrus sui langgraph msgpack
drop-in replacement

Three lines to decentralised memory

Replace SqliteSaver with WalrusCheckpointer. That's it. Your agent's memory is now stored on Walrus and registered on the Sui blockchain.

agent.py
from memwal.checkpoint import WalrusCheckpointer

# One line — loads config from .env
checkpointer = WalrusCheckpointer.from_env()

# Drop-in replacement for SqliteSaver
graph = builder.compile(checkpointer=checkpointer)

# Agent memory is now on Walrus + Sui
result = graph.invoke(
    {"messages": [HumanMessage(content="Hello, remember me!")]},
    config={"configurable": {"thread_id": "agent-001"}}
)
capabilities

Everything you need

A complete decentralised memory layer for LangGraph agents.

Persistent Memory

Checkpoint data survives across sessions, machines, and deployments. Stored on Walrus, registered on Sui.

Blob Storage

Store arbitrary binary data as Walrus blobs. Automatic retry with exponential backoff on transient failures.

Memory Retrieval

Fetch checkpoints by thread_id via the Sui on-chain registry. Any agent instance can resume any thread.

Agent Context

Thread-scoped checkpoints with namespace support. Multi-agent systems share a single registry.

On-Chain Registry

Thread→blob mappings stored as Sui Move dynamic fields. Verifiable, transparent, immutable.

API Playground

Interactive testing environment. Send requests, inspect responses, explore memory state in real-time.

architecture

How it works

From agent input to decentralised storage and back.

Input
Process
MemWal
Walrus
Sui
Output
applications

Built for builders

From personal assistants to multi-agent orchestration systems.

AI Agents

Give agents persistent, verifiable memory that survives across sessions and deployments.

Personal Assistants

Build assistants that remember conversations, preferences, and context permanently.

Multi-Agent Systems

Coordinate agents through shared on-chain state. Any agent can resume any thread.

Knowledge Workers

Accumulate and retrieve structured knowledge over time with checkpoint versioning.

Research Agents

Long-running research tasks with durable state. Resume from any checkpoint.

Long-term Memory

Decentralised, permanent storage for agent memory. No single point of failure.

get started

Try MemWal now

Open the interactive playground to test every API endpoint, explore memory state, and build with templates.

Open Playground