Decentralized persistent memory for AI agents.

One memory layer for every agent runtime.

Every memory your agent writes is an encrypted blob on Walrus. Every action it takes is a Merkle-chained attestation on Sui. Open one dashboard — across Claude Code, Hermes, Cursor and more — and search, replay, and prove exactly what it did.

Walrusencrypted at rest
Suimerkle-chained
8+runtimes
TraceSession
0x7a3f9c2e…d201
10 calls
plan_payment
coingecko.get_price
wallet.resolve_recipient
wallet.check_balance
Ready to verifyroot 0xb91e…f7a3
Built on the Sui · Walrus · Seal stack
SuiWalrusSealMemWalSui Overflow 2026
The problem

Agents are powerful, stateless, and impossible to audit.

Your agent's useful context gets scattered across tools, logs, and local machines. OneMem gives that context a portable memory namespace with a verifiable history.

Memory is trapped

Every app keeps its own silo. Switch tools and your agent forgets everything — there's no portable, shared memory.

Context loses its history

"It sent money / made a video / did research" - but which skill, which model, what inputs? The trace is gone the moment the run ends.

Nothing is provable

No way to show a teammate, an auditor, or a skeptic that the agent did exactly what it claims - untampered.

How it works

Three steps to persistent agent memory.

1

Install

One line adds OneMem to your agent runtime. No rewrites, no SDK gymnastics.

$ npm create onemem@latest
2

Use your agent normally

Memories and every tool / skill / MCP call are captured automatically - encrypted and chained as they happen.

agent.run("send 5 USDC to Maya")
3

Verify & share

Open the dashboard or CLI, walk the chain, and watch it turn green. Share a public proof link with anyone.

$ onemem verify 0x7a3f…d201
The pillars

Not one feature — a layer under every agent.

01memory

Persistent memory

Encrypted Walrus blobs, threshold-encrypted with Seal, access-controlled by on-chain namespaces. Cross-device sync without trusting a vendor.

02trace

Action trace + replay

Every tool / skill / MCP call captured as a Merkle-chained node. Replay any run purely from chain + Walrus — no original runtime needed.

03everywhere

Cross-runtime

The same memory namespace and trace format across every runtime and framework. One dashboard for all of them.

The signature moment

Proof turns the page green.

Click Verify and OneMem walks every call, recomputes each hash, and compares it to the on-chain root. Memory stays encrypted; integrity is public.

TraceSession
0x7a3f9c2e…d201
10 calls
plan_payment
coingecko.get_price
wallet.resolve_recipient
wallet.check_balance
Ready to verifyroot 0xb91e…f7a3
Integrations

Works where your agents already run.

Claude Code
Hermes
Cursor
OpenClaw
Vercel AI SDK
OpenAI Agents
CrewAI
LiveKit
ElevenLabs
Get started

Give every agent a memory layer.

Install in under a minute. Store a memory, run a trace, and prove it.