Your OpenClaw agent forgets everything between sessions. You already know this because you’ve dealt with the workarounds: MEMORY.md files, daily logs, manual context loading. They work until they don’t. Files get stale, context windows fill up, and your agent keeps re-learning things you told it last week.

MemoClaw gives your agent persistent memory through semantic vector search. Store things, recall them later by meaning rather than exact keywords. This guide gets you from nothing to a working memory setup in about 5 minutes.

What you need

• An OpenClaw instance (you probably have this if you’re reading this)
• The MemoClaw skill installed

That’s it. No API keys, no accounts. MemoClaw uses your wallet as your identity.

Step 1: Install the skill

clawhub install anajuliabit/memoclaw

That’s it. The skill lands in your workspace’s skills directory.

Step 2: Initialize MemoClaw

memoclaw init

This generates a wallet for your agent and saves the config. One command, done. Your wallet address becomes your identity — no accounts to create, no API keys to rotate.

If you already have an Ethereum wallet you’d rather use, memoclaw init lets you import an existing key too.

Step 3: Store your first memory

memoclaw store "Ana prefers direct communication, no fluff" --importance 0.8 --tags preferences,communication

That’s a real memory now. It lives on MemoClaw’s servers, indexed for semantic search. The importance score (0.0 to 1.0) helps with retrieval ranking. Tags let you filter later.

Try a few more:

memoclaw store "Project deadline is March 15 for the API v2 release" --importance 0.9 --tags deadlines,api

memoclaw store "Always run the linter before opening a PR on the API repo" --importance 0.7 --tags workflow,api

One thing to note: each memory has an 8192 character limit. This isn’t a file store. Think of it as individual facts, decisions, and observations, not entire documents.

Step 4: Recall memories

Here’s where it gets interesting. Recall uses semantic search, so you don’t need exact matches:

memoclaw recall "how does Ana like to communicate?"

This returns the memory about direct communication, even though you didn’t use those exact words in your query. The search is by meaning, not string matching.

memoclaw recall "upcoming deadlines"

Returns the March 15 API release memory.

memoclaw recall "PR process"

Returns the linter workflow memory.

Step 5: Wire it into your agent

The real value comes when your agent uses MemoClaw automatically. In your AGENTS.md, you can set up patterns like this:

## Memory

Use MemoClaw for persistent memory. NOT local markdown files.

### When to store
- User states a preference → importance 0.7-0.9
- User corrects you → importance 0.95
- Important decision made → importance 0.9
- Lesson learned → importance 0.8

### How
- Store: `memoclaw store "content" --importance 0.8 --tags tag1,tag2`
- Recall: `memoclaw recall "query"` before making assumptions
- Always recall before storing to avoid duplicates

With this in your AGENTS.md, your agent will recall relevant context before answering questions, store new information as it comes up, and build a growing knowledge base across sessions.

What happens in practice

Session 1, you tell your agent: “I use Neovim, not VS Code.”

The agent stores it:

memoclaw store "User uses Neovim, not VS Code" --importance 0.8 --tags preferences,editor

Session 47, you ask: “Set up my editor config.” Your agent recalls:

memoclaw recall "editor preferences"

And gets back the Neovim preference without you having to repeat yourself. It just knows.

A few things worth knowing

Recall before you store. If your agent stores “Ana likes coffee” and then later stores “Ana likes coffee in the morning,” you now have two overlapping memories. A quick recall first avoids duplication.

Importance scores matter for ranking. A memory stored at 0.95 ranks higher than one at 0.5 when both match a query. Use higher scores for corrections and hard facts, lower scores for casual observations.

Tags help with filtering. You can recall within specific tags:

memoclaw recall "deployment process" --tags infrastructure

This narrows the search to infrastructure-tagged memories only.

Namespaces separate contexts. If you run multiple projects or agents, use namespaces to keep memories isolated:

memoclaw store "API uses rate limiting of 100 req/min" --namespace project-alpha --importance 0.7

Free tier gives you 100 calls. That’s a mix of stores and recalls — enough to try everything in this guide without spending a cent. After that, MemoClaw uses x402 micropayments with USDC on Base. Each store or recall costs $0.005 USDC — that’s 200 operations per dollar.

What MemoClaw isn’t

It’s not a database you query with SQL. It’s not a file store (8192 char max per memory). It’s not self-hostable. It’s not a RAG pipeline, though you could build one on top of it. And don’t store secrets in it. Use a proper secrets manager for that.

What to do next

You have a working memory system. Some things to try:

• Add the recall/store patterns to your AGENTS.md so your agent uses them automatically
• If you have an existing MEMORY.md, migrate it over (we have a guide for that)
• Experiment with importance scores and tags to see how they affect recall quality

The docs at docs.memoclaw.com cover batch storage, immutable memories, and the REST API if you want to go deeper.

Your agent remembers things now. That’s the whole point.