Skip to content
Persistent memory for OpenClaw agents

Your OpenClaw agent
forgets nothing.

MEMORY.md files eat context and lose memories between sessions. MemoClaw gives your agent persistent semantic memory with one command — saving 80–95% of context tokens. No signup — just a wallet and 100 free calls.

<200ms recall99.9% uptimeSemantic search
Works withOpenClawClaude CodeCursorAny AgentMCP
memoclaw — demo

I'm an Agent

Get persistent memory for your OpenClaw agent in 30 seconds. Install, store, migrate.

1.clawhub install memoclaw
2.memoclaw store "User prefers dark mode"
3.memoclaw migrate ./MEMORY.md
Read SKILL.md →

I'm a Human

Learn how MemoClaw works, explore pricing, and see what makes it different.

See How It Works →

Or jump to pricing

Why teams switch to MemoClaw

🧠vs Mem0

No registration needed — wallet is your identity

vs Zep

No self-hosting, no infrastructure to manage

🤖vs Letta

Simpler API — store/recall, not a full agent framework

📄vs MEMORY.md

Semantic search, zero context tokens, 8× cheaper

Simple pricing

100 free calls to start. Then $0.005 per store/recall.
No subscriptions. No API keys. Just your wallet.

See full pricing

100

Free calls

per wallet

<200ms

Recall latency

p95 semantic search

$0.005

Per call

no minimums

4

Signal retrieval

hybrid scoring

Built for agents

How builders use MemoClaw

Replace MEMORY.md

Stop stuffing context into markdown files. Store memories semantically and recall only what's relevant — your agent's context window stays lean.

Zero-friction auth

No API keys, no signup forms. Your wallet is your identity. x402 means your agent just pays per call and it works — onboarding in seconds.

Semantic recall

Ask "what does the user prefer for deployments?" and MemoClaw surfaces the right memory from weeks ago. Grep can't do that.

Shared agent memory

Multiple agents sharing the same wallet can pass knowledge between sessions via namespaces. A shared brain for your agent fleet.

Why MemoClaw

Your OpenClaw agent wakes up blank every session. MEMORY.md files eat context windows and lose data between restarts. MemoClaw replaces flat files with semantic memory that actually scales.

Unique

Connection-Weighted Decay

Memories don't just fade with time — they strengthen when accessed together. Frequently recalled memories stay vivid. Isolated facts naturally decay.

Unique

4-Signal Hybrid Retrieval

Combines semantic similarity, recency, importance scoring, and access frequency. Not just vector search — intelligent memory that surfaces what matters most.

Proactive Suggestions

MemoClaw doesn't wait to be asked. It identifies stale high-importance memories, fresh unaccessed facts, and decaying knowledge that needs refresh.

Zero-Registration via x402

No signup forms, no API keys, no OAuth flows. Your wallet IS your identity. Pay-per-call micropayments mean you only pay for what you use.

Smart Ingestion

Dump raw conversations — MemoClaw extracts facts, deduplicates, scores importance, and auto-creates relations between memories.

MCP Native

First-class Model Context Protocol support. Drop one JSON config and your Claude Desktop, Cursor, or Windsurf gets persistent memory.

Privacy-First

Wallet-scoped data isolation, encrypted in transit, delete anytime. We never train on your data — your memories stay yours.

How it works

Two API calls. That's all your agent needs.

01

Store

Agent learns something important — a preference, decision, or correction. Store it with importance score and tags.

$
02

Recall

Query by meaning, not keywords. "What editor does the user like?" finds "prefers vim keybindings" automatically.

$
03

Auth

Sign with your wallet — 100 free calls. After that, x402 auto-payment kicks in. No API keys, no accounts.

$
Your Agent
MemoClaw
Persistent Memory

Memories persist across sessions, devices, and context window resets.

Quick Start

Install the CLI and give your agent persistent memory.

# Install the hook + CLI
npm install -g memoclaw
openclaw hooks install memoclaw-hooks
openclaw hooks enable memoclaw

# Set your private key
export MEMOCLAW_PRIVATE_KEY=0xYourPrivateKey

# Restart gateway — your agent now has persistent memory
openclaw gateway restart

See the FAQ for common questions. Using OpenClaw? Get the skill on ClawHub (opens in new tab) or skills.sh (opens in new tab).

New

Still using MEMORY.md?
One command. Zero data loss.

Import all your file-based memories — preferences, corrections, project context — into MemoClaw with a single command. Semantic search, decay scoring, and proactive suggestions included.

memoclaw migrate
# Tired of MEMORY.md eating your context window?
memoclaw migrate --from markdown

# That's it. All your memories, preferences, and context
# are now in MemoClaw with semantic search and decay scoring.

# Migrating from: MEMORY.md (found 847 memories)
████████████████████████████ 100%
✓ 847 memories imported
✓ 12 duplicates merged
✓ Relations auto-detected: 94

Zero downtime

Migrate while your agents keep running

🔗

Relations preserved

Memory connections transfer automatically

🔒

Same wallet

Your existing wallet works — no new accounts

Built for agents

MemoClaw solves the context problem for any AI agent. See detailed examples →

Personal Assistants

Remember user preferences, past conversations, and personal context across sessions.

Coding Agents

Store project architecture, tech decisions, and code style preferences. Never ask twice.

Research Agents

Build knowledge bases from research sessions. Recall findings by topic, not file name.

Multi-Agent Systems

Share context between agents. One agent learns, all agents remember.

Your data stays yours

Privacy-first design. No accounts, no tracking, no training.

Wallet-Scoped

Your memories are tied to your wallet. No one else can access them.

No Training

We never train on your data. Your memories stay yours.

Delete Anytime

Full control. Delete individual memories or everything at once.

Encrypted

All data encrypted in transit (TLS) and at rest (AES-256).

Built for real workloads

Numbers that matter when choosing a memory service for your agents.

<200ms

Average recall latency

Semantic search across thousands of memories

$0.005

Per store or recall

No subscriptions, no minimums, no hidden fees

0

Context tokens consumed

Unlike MEMORY.md, memories stay out of your prompt

30s

Setup time

memoclaw init → store → recall. That's it.

100

Free calls per wallet

No signup, no email, no credit card required

99.9%

Uptime

Running on Railway with automatic failover

What developers are saying

Replaced my 800-line MEMORY.md with two API calls. Context window usage dropped to zero overnight.

Marcus R.

AI agent developer

The x402 flow is genius — no API keys, no signup form, just pay and go. My agents were storing memories within a minute.

Priya K.

OpenClaw power user

We use namespaces to isolate per-project context. Semantic recall beats grep every single time.

Daniel S.

Full-stack engineer

Pricing

Only endpoints that use AI (embeddings + LLM) cost money. Everything else is free.

Free Tier

100 free calls

Every wallet gets 100 paid-endpoint calls free. Just sign with your wallet — no payment needed.

After free tier: pay-per-call with x402. Free endpoints are always free.

Store + recall for $0.005 each

No subscriptions. No API keys. No monthly minimums.

Paid endpoints

These use OpenAI embeddings or GPT-4o-mini

POST /v1/store$0.005

Store a memory

POST /v1/store/batch$0.04

Store up to 100

POST /v1/recall$0.005

Semantic search

PATCH /v1/memories/:id$0.005

Update a memory

PATCH /v1/memories/batch$0.005

Batch update

POST /v1/memories/extract$0.01

Extract facts via LLM

POST /v1/memories/consolidate$0.01

Merge similar memories

POST /v1/ingest$0.01

Auto-extract & relate

POST /v1/context$0.01

Assemble context block

POST /v1/migrate$0.01

Import markdown files

Free endpoints

No OpenAI cost — always free, no credits consumed

GET /v1/memoriesGET /v1/memories/:idDELETE /v1/memories/:idDELETE /v1/memoriesPOST /v1/searchGET /v1/suggestedGET /v1/memories/corePOST /v1/memories/coreGET /v1/memories/:id/historyPOST /v1/memories/:id/relationsGET /v1/memories/:id/relationsGET /v1/memories/:id/graphGET /v1/memories/exportGET /v1/namespacesGET /v1/stats

FAQ

Common questions about MemoClaw.

x402 is a payment protocol where paying for a request IS the authentication. Your wallet address becomes your identity. No signup, no API keys to manage, no rate limits to worry about. Just pay and use.

Local files require keyword search (grep), eat context window space when loaded, and don't work across devices. MemoClaw uses semantic search (find by meaning), persists outside your context, and works everywhere your wallet does.

USDC on Base (an Ethereum L2). Any EVM wallet works — MetaMask, Coinbase Wallet, Rainbow, etc. Base has low fees (~$0.01 per transaction).

Yes. Memories are scoped to your wallet address. Wallet A cannot see Wallet B's memories. We don't train on your data. You can delete memories anytime.

Any agent that can make HTTP calls or run shell commands. Claude Code, Cursor, Copilot, custom agents — if it can call an API, it works.

We generate embeddings (vector representations) of your memories using OpenAI. When you recall, we find memories with similar meaning — not just matching keywords. "editor preferences" finds "uses vim keybindings".

Paid endpoints will return an x402 payment-required response. Your agent (or SDK) automatically handles the micropayment with USDC on Base — typically under $0.01 per transaction. Free endpoints (list, delete, search, export, etc.) remain free forever.

There's no cap on the number of memories per wallet. Each individual memory can be up to 8,192 characters. For larger content, use the extract or ingest endpoints to break it into meaningful chunks automatically.

Yes. The GET /v1/memories/export endpoint returns all your memories as JSON — completely free, no credits consumed. You own your data and can take it with you anytime.

A typical MEMORY.md file consumes 2,000–10,000 tokens of your context window every prompt. With MemoClaw, your agent recalls only what's relevant — usually 200–500 tokens. That's an 80–95% reduction in memory-related context usage.

Yes. Any agent using the same wallet address can read and write to the same memory store. Use namespaces to organize memories by project or agent. It's like giving your agents a shared brain.

One command away from agents that remember.

memoclaw init — no signup, no email, no API keys. Your agent has persistent memory in 30 seconds.

$0.005 per store/recall · No monthly fees · Pay only for what you use