How to give your agent memory that survives restarts

If you've been using OpenClaw and noticed your agent forgets everything between sessions, you're missing one of the most useful features in the platform: persistent memory.

Out of the box, every conversation starts fresh. The agent doesn't remember your project, your preferences, or what you discussed yesterday. For one-off tasks that's fine. For an agent you actually rely on, it's painful.

Here's how to fix it in about 5 minutes.

Step 1 — Find your MEMORY.md file

Every OpenClaw install has a MEMORY.md file in its config directory. This is where persistent agent knowledge lives across sessions. If you've never edited it, it's probably empty.

Step 2 — Add the things your agent should always know

Open MEMORY.md and write things like:

Who you are and what you do
- Projects or clients the agent helps with
- Tone and writing style preferences
- Tools, accounts, or services it should know about

Plain English works fine. No special syntax needed.

Step 3 — Use it

Next time you start a session, the agent reads MEMORY.md automatically. Ask it something it should now "remember" — like "what's my tone of voice for client work?" — and you'll see the difference instantly.

Step 4 — Update it as you go

When you tell your agent something useful, ask it to add that fact to MEMORY.md. Over time you build up an agent that actually knows you, instead of one that resets every morning.

This is the single biggest upgrade most people miss. What's the first thing you'd put in your MEMORY.md? Drop it below.

2 comments