A no-jargon guide for designers, creatives, and non-developers who want to get the most out of their AI agent.
No developer background requiredOpenClaw is software you install on your own computer (or a server) that turns it into a personal AI assistant hub. Once it's running, you can talk to your AI from any messaging app — WhatsApp, Telegram, Discord, iMessage, Slack — and it replies with full awareness of your projects, files, and history.
Think of it like this: instead of having a separate AI tab open in Chrome that forgets everything when you close it, OpenClaw runs in the background on your machine and is always available, from anywhere, with memory of everything you've told it.
The lobster is the mascot. Don't worry about it.
The Gateway is the engine. It runs silently in the background and handles all the connections. You never have to think about it once it's set up.
OpenClaw stores everything in a folder called the workspace. By default it lives at ~/.openclaw/workspace/ — the ~ just means your home folder. The .openclaw part is hidden in Finder by default — press Cmd+Shift+. to reveal hidden files on a Mac.
Here's what each file does:
This defines the personality and role of your agent. Think of it like a job description combined with a personality sheet.
This tells the agent about you. The more it knows, the better it can help.
The operating manual for your agent. It loads every single session — think of it as standing orders.
The agent's permanent brain. Things written here persist across every conversation, every session, forever — until you edit them out.
Daily diary entries — one file per day, auto-created. The agent reads today's and yesterday's logs at the start of every session, so it has recent context without you re-explaining everything.
A notes file for things specific to your machine — not personality or memory, just practical reference information.
If you want the agent to check something on a schedule (monitoring a service, checking for emails, etc.), put those tasks here. Most people leave this empty — which is completely fine.
A simple card with the agent's name, emoji, and vibe. Used in dashboards and UI displays. Rarely needs editing.
There's one more important file that lives outside the workspace:
~/.openclaw/openclaw.jsonThis is the master control file for OpenClaw itself. It controls things like which AI model to use, which messaging apps are connected, who is allowed to message the agent, security settings, and multi-agent setup.
You usually don't edit this by hand. OpenClaw has commands and a web UI for most settings. But knowing it exists is useful if something isn't behaving right.
Skills are like plugins for your agent. Each skill teaches it how to do something specific — interact with Apple Notes, control Slack, send WhatsApp messages, check the weather, generate images, and more.
Each skill is a folder with a SKILL.md file inside it. When you ask the agent to do something that matches a skill's description, it automatically reads those instructions and follows them. You don't activate skills manually — the agent figures out when to use them.
To install a community skill:
openclaw skills install <skill-name>Browse available skills at clawhub.ai
You can run more than one agent on the same Gateway. Each agent has its own workspace, personality, memory, and chat history. Messages get routed to the right agent automatically.
For example: one agent for work, one for personal use, one specialized for code — all running side by side, never getting confused with each other.
To add a new agent:
openclaw agents add myagentOpenClaw connects to messaging apps called channels. Each channel requires a one-time setup, but once it's done, the agent is available there permanently.
OpenClaw has a built-in web dashboard you can open in any browser:
http://localhost:18789From here you can chat with your agent directly, see all active sessions, view and edit config, check which channels are connected, and monitor agent activity.
The more specific your instructions, the better. Instead of "be helpful," write exactly what done looks like — what to report back, what to never do, how to handle errors.
If the agent is too formal or too casual, fix it here. Small changes make a big difference in how natural conversations feel.
If something comes up in conversation that should stick permanently, just say "remember that." It'll write it to MEMORY.md automatically.
Add a section to AGENTS.md for each major project — what it is, where the files live, what tools are involved. The agent will stop asking obvious questions.
Over time it gets cluttered. Review it occasionally and remove outdated info. Think of it like cleaning out a filing cabinet.
If you repeatedly ask the agent to do something, install the relevant skill. It'll do it faster and more reliably with proper instructions.
OpenClaw installs as a background service that starts automatically. You shouldn't need to do anything — it's just there.
Your workspace files and conversation history stay on your machine. The only thing that leaves your computer is the messages you send to the AI model provider (Anthropic, OpenAI, etc.) — same as using their apps directly.
Yes. OpenClaw supports Ollama for running models like Qwen, Llama, etc. entirely on your machine with zero API costs.
npm install -g openclaw@latest
openclaw gateway restartopenclaw doctorThis runs a diagnostic check and tells you exactly what's wrong.
| File | Purpose | Should You Edit It? |
|---|---|---|
| SOUL.md | Agent personality & role | Yes — define who it is |
| USER.md | About you | Yes — tell it about yourself |
| AGENTS.md | Standing rules & instructions | Yes — most powerful file |
| MEMORY.md | Long-term memory | Yes, carefully |
| memory/YYYY-MM-DD.md | Daily logs | Agent writes, you can read |
| TOOLS.md | Your local setup notes | Yes — your reference doc |
| HEARTBEAT.md | Scheduled tasks | Only if you want them |
| IDENTITY.md | Agent name/emoji | Rarely |
| ~/.openclaw/openclaw.json | Master config | Via UI or CLI |