Give your AI agent a real memory
in 10 minutes

Most AI assistants forget everything the moment a conversation ends. They can't remember what you told them last week, can't notice when two things they know contradict each other, and can't explain why they believe something.

Mnemebrain changes that. It gives your AI agent a structured belief memory — not just a list of facts, but a system that tracks evidence, detects conflicts, and learns over time. And AI Skills for OpenClaw let you drop that memory into any AI workflow without writing a single line of code.

This guide walks you through the whole thing, step by step, like you've never touched a terminal before (even if you have).

Think of a normal AI assistant's memory like a sticky note. You write something on it, and later it can read it back. But it doesn't know if that note is still true, doesn't notice if two sticky notes say opposite things, and has no idea where the information came from.

Mnemebrain replaces the sticky note with something closer to how your brain actually works: every piece of knowledge has evidence (where did this come from?), a confidence level (how sure are we?), and a history (has this ever changed?).

When your agent learns something new that contradicts an old belief, Mnemebrain notices, flags the conflict, and updates accordingly — the same way a thoughtful person would.

Mnemebrain Lite is the open-source, local version. It runs entirely on your machine, stores everything in a local database, and is free to use. The full version (accessible at localhost:8000 when running) adds more advanced features — but Lite is the perfect place to start.

First, open your Terminal (on Mac: press ⌘ + Space, type "Terminal", press Enter) or Command Prompt (on Windows: press the Windows key, type "cmd", press Enter).

Check that Python is installed by typing this and pressing Enter:

python --version

You should see something like Python 3.10.4. Any version 3.10 or higher works. If you see an error, download Python here first.

Now install Mnemebrain Lite:

pip install mnemebrain-lite

That's it. pip (Python's package installer) downloads and installs everything automatically. You'll see a progress bar, then a success message.

Confirm it installed correctly:

python -c "import mnemebrain_core; print('Mnemebrain ready ✓')"

If you see Mnemebrain ready ✓ — you're set.

Mnemebrain runs as a small local server on your computer. Your AI agent connects to it to store and retrieve beliefs. Start it with:

mnemebrain-lite serve

You should see output like:

INFO     MnemeBrain Lite v0.x.x starting...
INFO     Storage: ./mnemebrain.db (Kuzu graph database)
INFO     Server running on http://localhost:8000
INFO     API docs:  http://localhost:8000/docs

Leave this terminal window open. The server needs to keep running. Open a new terminal window for the next steps.

OpenClaw uses "AI Skills" — pre-built modules you can drop into any agent to give it new abilities. The Mnemebrain skill pack teaches your agent how to read from and write to its belief memory.

Download the skills from GitHub. In your terminal (the new one you opened), run:

# Download the Mnemebrain skills repository
git clone https://github.com/mnemebrain/mnemebrain-skills.git

# Move into the folder
cd mnemebrain-skills

If you don't have git, you can also download the ZIP file directly from GitHub and unzip it anywhere on your computer.

Inside, you'll find two main skill folders:

Now install the integration skill's dependencies:

cd mnemebrain-lite-integration
pip install -r requirements.txt

The skill needs to know where your Mnemebrain server is running. Open the file called config.yaml inside the mnemebrain-lite-integration folder. You can use any text editor — Notepad, TextEdit, VS Code, whatever you have.

It will look something like this:

mnemebrain:
  host: http://localhost:8000    # leave this as-is
  agent_id: my-agent             # give your agent a name
  auto_store: true               # store beliefs automatically

Change my-agent to whatever name you want for your agent (no spaces — use hyphens like project-helper or sales-bot). Save the file.

Now register the skill with OpenClaw. In OpenClaw's interface, go to Settings → Skills → Add Skill and point it to the mnemebrain-lite-integration folder you just configured.

Start a conversation with your agent in OpenClaw. Tell it something you want it to remember:

My budget for this project is $5,000 and the deadline is April 15th.

Behind the scenes, Mnemebrain just stored two beliefs:

{
  "claim":      "project budget is $5,000",
  "confidence": 0.95,
  "evidence":   ["user_message_001"],
  "source":     "user_direct_statement"
}

{
  "claim":      "project deadline is April 15th",
  "confidence": 0.95,
  "evidence":   ["user_message_001"],
  "source":     "user_direct_statement"
}

Now close the conversation. Start a completely new one. Ask:

What's our project deadline?

Your agent will answer: "Your project deadline is April 15th." — without you having to repeat yourself.

Now try something more interesting. Tell it something that contradicts the first thing:

Actually, the budget was just increased to $8,000.

This is what separates Mnemebrain from a simple memory store: it doesn't just overwrite. It reasons. No other memory tool for OpenClaw does this — a plain vector store would silently replace the old fact and you'd never know a conflict existed.

You can inspect everything your agent believes at any time. The easiest way is the interactive API explorer — open your browser and go to:

http://localhost:8000/docs

You'll see a list of all the available operations. Click GET /beliefs, then click Try it out → Execute to see everything currently in memory.

You can also run the proof skill to verify everything is connected and working:

# From the mnemebrain-skills folder
cd mnemebrain-lite-proof
python run_proof.py

This runs a quick end-to-end test: it stores a belief, retrieves it, creates a contradiction, verifies the conflict is detected, then cleans up. If everything passes, you'll see:

[1/4] Storing belief...          ✓
[2/4] Retrieving belief...        ✓
[3/4] Detecting contradiction...  ✓
[4/4] Verifying revision...       ✓

All checks passed. Mnemebrain is working correctly.

That's it. Your AI agent now has a real, structured memory that persists between conversations, detects contradictions, and can explain what it knows and why.