How to Set Up OpenClaw with Pieces Long-Term Memory

If you're using AI agents, one of the biggest limitations is memory. In this guide, I'll show you how to connect OpenClaw to Pieces Long-Term Memory so your agent instantly knows everything about you and your work, and can operate 24/7 while building context from everything it does.

What You'll Need:

1. A computer with Node.js installed

2. Pieces Desktop App

3. ngrok

4. An OpenClaw-compatible API key

Setup Time: 3-5 minutes

Step 1: Download Pieces and Enable Long-Term Memory

Download Pieces from pieces.app and install the desktop app.

Once installed, open Pieces and go to Settings → Long-Term Memory and make sure it's enabled.

This allows Pieces to capture your real work — your screen, clipboard, and activity — and turn it into structured memory over time.

The more you use your computer, the richer your memory gets.

Step 2: Start the Tunnel

To make your local Pieces instance accessible to OpenClaw, you'll need to expose it using ngrok. In your terminal, run:

ngrok http 39300

You'll get a URL that looks like this:

https://abc123.ngrok-free.dev

Example of terminal output:

Copy that URL — you'll need it in a later step. Keep this terminal window running.

Step 3: Install OpenClaw

Head to https://openclaw.ai and copy the install command from the site.

Paste it into your terminal and run it.

Step 4: Select Your Model and Launch OpenClaw

When prompted, choose your model and add your API key.

Then launch OpenClaw in the browser:

openclaw dashboard

You should now see the OpenClaw interface open in your browser.

Step 5: Install the Pieces Long-Term Memory Skill

In OpenClaw, navigate to the Skills page and click Browse.

Search for Pieces and install Pieces Long-Term Memory MCP.

Or download it directly at: https://clawhub.ai/jackrosspieces/pieces-mcp

This skill teaches your agent how to connect to Pieces, query your long-term memory, and create new memories.

Step 6: Let the Agent Configure Itself

Go to the chat and type:

"Set up my Pieces long-term memory"

OpenClaw will read the skill and begin configuring everything automatically.

When prompted, paste the ngrok URL you copied in Step 2.

From there, the agent handles the rest — building the MCP endpoint, configuring MCPorter, installing dependencies, and restarting the gateway.

Step 7: Test It

After your gateway restarts, ask your agent:

It will query Pieces Long-Term Memory and return an answer based on your real activity — not guesses, not generic responses. Your actual work.

Step 8: Create a Memory

Ask your agent:

That memory gets written directly into Pieces. Open the Pieces app and you'll see it immediately. It's now part of your long-term context, available in every future session.

What This Unlocks

Once connected, your OpenClaw agent can:

1. Recall what you've worked on across sessions

2. Understand your projects and past decisions

3. Reference your history without you re-explaining anything

4. Build new memories automatically over time

Your agent isn't starting from scratch anymore — it already knows you.

You can use this same skill.md file to teach any AI agent how to connect and use Pieces as it's long-term memory!