What are the five components of OpenClaw?

Gateway, Agent, Channels, Skills, and Memory. The Gateway is the always-running process that routes messages. The Agent is the AI that processes and responds. Channels are the messaging apps you connect (Telegram, Slack, etc.). Skills are plugins that give the agent new abilities. Memory is the set of files (SOUL.md, USER.md, MEMORY.md) that persist context across conversations.

Does OpenClaw send my data to the cloud?

Only the message content goes to your AI provider (e.g., Anthropic or OpenAI) to generate a response. Your files, memory, and configuration stay on your machine. If you run a local model (Ollama), nothing leaves your device.

What is the difference between a local and cloud AI model in OpenClaw?

Cloud models (Claude, GPT-4) send your messages to an external API. Local models (via Ollama) run entirely on your hardware with no network calls. Cloud models are faster and more capable; local models are fully private and work offline.

Module 1: How Everything Fits Together

Understand how OpenClaw works in 5 minutes. A plain-English guide to the Gateway, channels, skills, and AI provider — no technical background needed.

0 of 10 modules complete0%

5 min read

What you will learn

Name the five core components of OpenClaw
Trace how a message flows from your phone to an action
Understand the difference between local and cloud models
Know where your data lives and who controls it

Why this matters

Before building anything complex, it helps to see the full picture. OpenClaw has five simple parts. Once you know what they are and how they connect, every future module will make immediate sense.

The five components

Component	What it does	Where it lives
Gateway	Receives messages from all your channels	Background process on your machine
Agent	Decides what to do with each message	Core logic layer
Memory	Stores personality, preferences, and context	Markdown files in your workspace
Skills	Gives the agent abilities (email, calendar, search)	Installed via ClawHub
AI Model	Does the actual thinking	Cloud (OpenAI/Claude) or local (Ollama)

How a message flows

When you send "What's on my calendar today?" via WhatsApp:

The whole round trip takes 2-5 seconds depending on your model and network.

Local vs cloud models

You choose where the "thinking" happens:

Option	Speed	Privacy	Cost
OpenAI (GPT-4o)	Fast	Data sent to OpenAI servers	Pay per use
Anthropic (Claude)	Fast	Data sent to Anthropic servers	Pay per use
Ollama (local)	Depends on hardware	Everything stays on your machine	Free

You can switch models at any time. Your data, memory files, and skills are completely independent of which model you use.

Can I use multiple models?

Yes. You can configure different models for different tasks. For example, use a fast local model for simple questions and a cloud model for complex reasoning.

Where your data lives

Everything OpenClaw stores is in your workspace directory:

~/.openclaw/
├── SOUL.md          # Agent personality and behavior rules
├── USER.md          # Information about you
├── MEMORY.md        # Things the agent remembers across conversations
├── HEARTBEAT.md     # Standing context and priorities
├── workspace/       # Files the agent can read and write
├── skills/          # Installed skills
├── tokens/          # OAuth tokens (don't edit these)
└── openclaw.json    # All configuration (API keys, channels, settings)

The ~ symbol means your home folder — on macOS it is /Users/yourname, on Linux /home/yourname, on Windows C:\Users\yourname.

No data is stored on external servers (unless you choose a cloud AI model, which only sees the messages you send it). You own everything.

Hands-on: Check your components

Run the health check

openclaw doctor

You should see output like this — green checks mean everything is working:

✓ Node.js         v24.0.1
✓ OpenClaw        v2026.3.24
✓ Gateway         running
✓ Agent           ready
✓ Model           gpt-4o (OpenAI) — connected
✓ Config          valid

If anything shows a red ✗, the output will tell you what to fix.

Find your workspace

openclaw config get workspace

This prints the path to your agent's files — SOUL.md, USER.md, MEMORY.md, and your config. Open that folder in Finder (macOS) or File Explorer (Windows) to explore.

Read the reference

For precise definitions of Gateway, Agent, SOUL.md, and every other OpenClaw concept, see the Key Concepts reference.

Right now you can only talk to your agent through the Terminal. In the next module you will connect it to your phone so you can message it like a regular app — no Terminal needed after that.

Finished this module?

Tracks your progress across all 10 modules

FAQ

What are the five components of OpenClaw?: Gateway, Agent, Channels, Skills, and Memory. The Gateway is the always-running process that routes messages. The Agent is the AI that processes and responds. Channels are the messaging apps you connect (Telegram, Slack, etc.). Skills are plugins that give the agent new abilities. Memory is the set of files (SOUL.md, USER.md, MEMORY.md) that persist context across conversations.
Does OpenClaw send my data to the cloud?: Only the message content goes to your AI provider (e.g., Anthropic or OpenAI) to generate a response. Your files, memory, and configuration stay on your machine. If you run a local model (Ollama), nothing leaves your device.
What is the difference between a local and cloud AI model in OpenClaw?: Cloud models (Claude, GPT-4) send your messages to an external API. Local models (via Ollama) run entirely on your hardware with no network calls. Cloud models are faster and more capable; local models are fully private and work offline.

Module 1: How Everything Fits Together

On this page