Teaching OpenClaw How You Work

OpenClaw gets more useful the more it knows about you. Out of the box it knows nothing: it doesn't know your name, your job, your projects, or how you like things done. Every session starts fresh.

Persistent memory changes that. Once it is on, OpenClaw builds up a picture of how you work over time. This tutorial is about doing that deliberately, not just waiting for it to happen accidentally.

What persistent memory actually stores

When memory is enabled, OpenClaw saves things it learns about you between sessions. This includes:

• Preferences you state or that it picks up from how you work
• Context about ongoing projects
• Your communication style and formatting preferences
• Tools, folders, and apps you use regularly
• Corrections you make when it does something wrong

The memory is stored locally on your computer. You can view it, edit it, and delete entries at any time.

Why it matters

Without memory, every session starts from zero. You have to re-explain your context every time. With memory, you can say "start a summary for the Henderson project" and OpenClaw already knows what that project is, who the key people are, and how you like summaries formatted.

This sounds small but it compounds quickly. After a few weeks of regular use, the difference between memory-on and memory-off is the difference between a tool that knows your workflow and a tool you have to babysit.

Step 1: Tell it the basics

Start a conversation and explicitly introduce yourself. This is faster than waiting for it to pick things up organically.

Here is a good template to adapt:

I want to set up my profile so you can work better for me over time.
Here is what you should know:

My name is [name]. I work as a [job title] at [company or type of org].

My main focus right now is [2 or 3 current projects or priorities].

Tools I use every day: [list your main apps, e.g. Notion, Gmail, VS Code, Slack]

My working style: I prefer [brief descriptions - e.g. concise summaries over long explanations,
bullet points over paragraphs, informal tone in drafts].

Folders I care about: [list key folder paths if relevant]

Please store all of this so you can reference it in future sessions.

After you send this, ask it to confirm: "What did you store from that?" Read through the response and correct anything it got wrong.

Step 2: Add project context

For each active project you want OpenClaw to know about, give it a brief. Keep it practical.

Please remember the following about a project called Henderson Rebrand:
- Client: Henderson Manufacturing, based in Chicago
- My main contact is Sarah Chen, sarah@hendersonmfg.com
- We are working on a website redesign and a new brand identity
- The deadline for the first deliverable is April 30
- All files for this project live in /Documents/Projects/Henderson/
- I prefer to name documents with the format YYYYMMDD_Henderson_filename

Do this for two or three active projects and you will notice immediately how much smoother tasks feel.

Step 3: Set formatting preferences

OpenClaw has default ways of formatting responses. If they don't match how you think, correct them explicitly.

For everything you produce for me, please follow these formatting rules:
- Summaries should be bullet points, not paragraphs
- Email drafts should use a casual but professional tone, not overly formal
- Lists should be numbered when order matters, bulleted otherwise
- Never use the word "utilize" - say "use" instead
- Keep outputs under 300 words unless I ask for something longer

These preferences get stored and applied going forward. You do not have to repeat them.

Step 4: Check what it has stored

At any point you can ask:

What do you currently know about me and my preferences?

Read through the response carefully. You are looking for:

• Anything that is just wrong
• Preferences it has stored that you no longer want
• Important context that is missing

For anything that needs fixing:

You have stored that I work in finance, but that is outdated. I moved
to product management 6 months ago. Please update that.

Remove the preference you stored about always using formal language.
I've changed how I want emails written.

Step 5: Correct it when it gets something wrong in a session

The most natural way to improve memory is to correct OpenClaw when it does something that doesn't match your preferences, and then explicitly ask it to remember the correction.

For example, if it formats a summary in a way you don't like:

That format doesn't work for me. I prefer shorter bullets with no sub-points.
Please redo the summary and store this preference for future sessions.

The key phrase is "store this preference for future sessions." Without that, it will just apply the correction now and forget it next time.

What good memory looks like in practice

After a few weeks of using OpenClaw with memory enabled and maintained, a session might start like this:

You: "Catch me up on what's happening with Henderson this week."

OpenClaw: "Here's a summary based on files in /Documents/Projects/Henderson/ from the past 7 days..."

It knows the project, it knows where the files live, and it knows you like summaries in bullets. You did not have to say any of that. That is what a well-built memory profile feels like.

A practical schedule for maintenance

You don't need to actively manage memory all the time. A good rhythm is:

• First week: Set up your profile, add 2 to 3 project contexts, set formatting preferences
• Ongoing: When it does something wrong, correct it and ask it to remember the correction
• Monthly: Ask "what do you know about me?" and do a 5-minute review and cleanup

That is all it takes to keep a memory profile useful without it becoming a chore.