What is OpenClaw
You spend every day inside some chat app. If there were a "contact" in it where you send "remind me about the meeting at 9 tomorrow" and it notes it down and reminds you on time; you send "keep an eye on this webpage for me" and it keeps watching. And this contact doesn't live on someone else's cloud — it lives on your own computer.
Send a message and try it, and see where that message goes.
Your chat app
Feishu, WeCom, Telegram…
The one you're already using
Your machine
- Sessions
- Memory
- Credentials
Model provider
DeepSeekSwappable pluginsClick a message in the chat app on the left and send it to your assistant.
OpenClaw is an open-source AI assistant that runs on your own devices. A process called Gateway stays resident on your computer or server, then waits for you inside the chat apps you already use.
This section doesn't install any software. After reading it you'll know what pieces it's made of, whether it costs money, and whether your chat app can connect.
What you'll be able to do after this section
- Explain in one sentence what OpenClaw is
- Say what the Gateway does and what stays on your machine
- Find out whether the chat app you use every day can connect
Three puzzle pieces
There were three things in the animation just now, and every step after installation revolves around them.
Entry points: the places you already use. Chat apps, the browser-based console (Control UI), the terminal (CLI / TUI), and native apps on each platform. They all connect to the same Gateway — they don't each go their own way.
Gateway: the process that lives on your machine. Officially it's called the "local control plane," and it manages four things: sessions, tools, events, and channel connections. Your state, memory, and credentials are all stored on your own hardware.
Model: the swappable plugin behind the Gateway. Switching models doesn't mean switching assistants. Chinese models like DeepSeek, Qwen, GLM, Kimi, MiniMax, and Doubao all have official support, and you can also connect local models like Ollama and vLLM. In the animation, switch to another model and send another message — the memory is still there.
Quiz timeYou switched the model from DeepSeek to Qwen. Are the reminders you previously asked the assistant to remember still there?
- I can name the four things the Gateway manages: sessions, tools, events, and channel connections
- I know the model is swappable, and memory doesn't follow the model
What it can do
| Capability | Description |
|---|---|
| Channels | 20+ chat channels connect to one Gateway; @mention it in groups, pair in direct messages |
| Tools | Browser automation, command execution, web search, sandboxing |
| Automation | Scheduled tasks, webhook triggers, email watch duty |
| Memory | Plain Markdown files plus a SQLite index, no hidden state, can recall across sessions |
| Media | Send and receive images, audio, video, documents; image generation, speech synthesis, speech-to-text |
| Multi-agent | One Gateway runs multiple mutually isolated agents |
The official FAQ lists five most common uses: personal briefings, research and first drafts, scheduled reminders and follow-ups, browser automation, and cross-device collaboration (send a task from your phone, run it on the server, come back to chat to get the result). The three lines in the animation are taken from three of them.
Can my chat app connect
For users in China, this is the first hurdle in deciding whether to use it.
Look it upWhich one do you usually use?
Click one to see.
How each channel connects is covered in scenario walkthroughs.
- I've checked the chat app I use every day
Does it cost money, and what about privacy
It's free. The whole product is MIT-licensed and governed by the independent nonprofit OpenClaw Foundation. Officially stated: no paid tier, no hosted service, no token. The only thing you pay for is the API fees of the model provider you choose, paid to the model vendor. If you connect a local model, even that goes away.
The data is in your hands, but you need to tell two things apart. State, memory, and credentials are stored on your hardware. But the messages you send ultimately have to go to the model provider and chat platform you've configured — no AI assistant can avoid that. If you want your conversation content to stay home as much as possible, pick a local model like in the animation; see local models.
OpenClaw itself only does a once-a-day version check to the foundation by default (with version number, OS, Node version, and CPU architecture); anonymous feature statistics require you to opt in. Setting update.checkOnStart: false turns all of it off.
- I know the core is free, and only the model API might cost money
- I know "data stays local" doesn't mean "messages never leave"
Compared with web-based AI
| Web-based / hosted AI | OpenClaw | |
|---|---|---|
| Where it runs | The platform's cloud | Your computer or server |
| Data | Conversation history on the platform | Workspace, sessions, credentials on your hardware |
| Cost | Subscription or pay-as-you-go | Core is free and open source; you only pay for the model API |
| Model | Whatever the platform gives you | Swappable plugins; domestic and local models both work |
| Entry point | A dedicated webpage or app | The chat apps and native apps you already have |
| Getting work done | Mostly conversation | Scheduled tasks, browser, command execution |
Checkpoints
- I can explain OpenClaw in one sentence: an open-source AI assistant that runs on my own devices and shows up in the chat apps I already use
- I know the Gateway is the center — all entry points connect to it, and models are swappable behind it
- I know the core is free under the MIT license, with no paid tier
- I've checked whether my chat app can connect
Common misconceptions
Assume it is an OpenAI product. It is not. It is managed by the independent OpenClaw Foundation, and OpenAI is just one of the donors.
Follow old tutorials for Clawdbot or Moltbot. These two are old names of the same project. The commands and configuration fields in the old tutorials have most likely stopped working. "OpenClaw 2.0" refers to v2026.8.1, and this course follows v2026.10.9.
Assume you need to buy an "enterprise edition". There is no such thing. The paid screenshots you find are most likely about some other product.
Treat it as a replacement for Claude Code. The two are complementary. The official positioning is: for writing code inside a code repository, using Claude Code or Codex directly is fastest; OpenClaw handles persistent memory, cross-device access, and tool orchestration.
FAQ
Does Clawdbot have a web version?
Clawdbot is the old name of this project; it is now called OpenClaw. It does not have a dedicated "web version" — the entry points are the chat apps you already use, the browser-based console (Control UI), the terminal (CLI / TUI), and native apps for various platforms, all of which connect to the same Gateway running on your own machine.
Can Clawdbot connect to DeepSeek?
Yes. DeepSeek is one of the officially supported models, and other Chinese models such as Qwen, GLM, Kimi, MiniMax, and Doubao all work too. A model is just a swappable plugin behind the Gateway — you can switch models without switching assistants, and your memory does not follow the model.
What is the relationship between OpenClaw and OpenAI?
It is not an OpenAI product. OpenClaw is governed by the independent nonprofit OpenClaw Foundation, and OpenAI is just one of its donors. The entire product is open source under the MIT license.
Can OpenClaw replace Claude Code?
The two are complementary, not a replacement. The official positioning is: when writing code inside a code repository, using Claude Code or Codex directly is fastest; OpenClaw handles persistent memory, cross-device access, and tool orchestration.
Next Section
OpenClaw 2.0 What's New, or go straight to Installation.