OpenClaw is an open-source personal AI assistant that runs on a computer you control and that you reach through messaging apps you already use.
OpenClaw is an open-source AI assistant you run on your own machine and talk to from chat apps you already use, such as WhatsApp, Telegram, Discord, Slack, Signal, or iMessage. It can keep memory and skills, work with models you bring, and take actions on the computer it runs on. There is no hosted subscription tier. The project is stewarded by the OpenClaw Foundation, an independent 501(c)(3), and the software is MIT licensed. We have worked with OpenClaw and set it up for clients. This page explains what it is, what it can do, and when it is a reasonable fit. Product facts are taken from the official site, openclaw.ai, and are current as of November 10, 2026.
What is OpenClaw?
Most people first meet AI as a chatbot in a browser tab. You type a question, you get an answer, and nothing else on your computer changes. OpenClaw is built as an assistant that can also do work: read and draft email, work with a calendar, browse the web, fill forms, read and write files, and run commands, depending on how it is configured. You message it in a chat app you already have, instead of opening a new tool every time.
It runs on macOS, Windows, or Linux. Official desktop apps install the gateway, chat, and setup pieces for you, and there is also a command-line installer. State — memories, skills, and configuration — lives on your machine, not on a vendor cloud that you subscribe to. You bring the model: hosted, subscription-backed, gateway, or local, as the official site describes it.
The OpenClaw Foundation employs the core team, signs releases, and keeps the product MIT licensed. The official site is explicit: there is no enterprise edition and no paid version, and a default install sends the Foundation nothing but a daily version check, which you can turn off. We treat that as the official policy, not as a sales claim of our own.
What can it do?
The official product pages describe a few capabilities that matter in day-to-day work:
- Chat where you already are. OpenClaw lists WhatsApp, Telegram, Discord, Slack, Signal, and iMessage among its channels, and says it works in DMs and group chats, with a broader set of 29 channels in total.
- Persistent memory. It can remember preferences, projects, and people from one conversation to the next, so you are not starting from zero every morning.
- Skills and plugins. You can extend it with community skills or build your own. The official site notes that it can also write its own skills.
- Access to the machine, with a choice of how far that goes. It can browse, fill forms, read and write files, and run shell commands. Official docs describe both full access and sandboxed setups, and they treat that choice as yours.
- Team use. You can share one gateway so a team can open and steer sessions. OpenClaw also publishes a Multiplayer product note for that kind of work.
Those are official capabilities, not a promise that every business should turn all of them on. In a real setup, we usually start narrower: one or two jobs, a limited set of tools, and a clear rule for when the assistant must stop and ask a person.
How do we set it up?
We have worked with OpenClaw and set it up for clients. The work is practical, not a product demo.
First we learn what you already use — email, calendar, chat, and any AI tools — and which job should come first. Inbox follow-up, appointment scheduling, and after-hours questions are common starting points. Then we recommend whether OpenClaw is the right tool, or whether something else fits better. We do not push a single platform.
If OpenClaw is the right fit, a typical setup includes installing it on a machine you control, connecting the chat channel you actually use, choosing a model you can pay for and operate, and limiting what the assistant is allowed to do. Official security guidance starts from conservative defaults: on a regular host install the gateway binds to loopback, unknown direct-message senders get a pairing code instead of having their message processed, and group access is allowlisted. We follow that guidance, run the project's own `openclaw security audit` where it applies, and only widen access when there is a reason.
We also set the human-in-the-loop rules before anything can message a customer or change a calendar hold. The discovery call is for diagnosis. Setup itself is paid work, quoted after that call, and we walk you through how to use and supervise the assistant when it is ready.
When is OpenClaw a good fit?
OpenClaw is often a good fit when you want the assistant to live on a computer you control, you already live in one of the chat apps it supports, and you do not want a hosted subscription for the assistant itself. It can also fit a small team that is willing to share one gateway and treat that gateway as a single trust boundary.
It is less of a fit if you want a fully hosted teammate with no machine to operate. In that case, a hosted option such as Grok Bot may be the better starting point. If you want a local open-source agent with an optional hosted cloud agent and a different official feature set, Hermes Agent is the comparison to make. We wrote a longer side-by-side in Hermes Agent vs OpenClaw.
OpenClaw is also not a shortcut past judgment. An assistant that can send email or run commands is useful only if someone decides what it may touch, what it must ask, and how you will notice when it is wrong.
How is security handled?
A useful assistant has access. That is the point, and also the risk. OpenClaw's official security docs start from conservative defaults and a simple model: one trust boundary per gateway — a single operator, or a team whose members trust each other. The project is explicit that it is not a hostile multi-tenant security boundary for mutually adversarial users sharing one agent. If you need that kind of isolation, official guidance is to split gateways and credentials, and ideally use separate OS users or hosts.
If you are evaluating OpenClaw, read docs.openclaw.ai/security and the project's security page. Our own overview of the basics — for OpenClaw and for other agents — is in Securing an AI agent: basics.
Official sources
Facts on this page are taken from those sources as of November 10, 2026. If a later official change conflicts with something here, the official site wins. We do not invent features, client names, or results.
Questions
Is OpenClaw free?
The software is open source under the MIT license. The official site says there is no subscription, no hosted tier, and no paid version. You still need a computer to run it on, and you bring a model, which may have its own cost. We quote our setup work separately, after a discovery call.
Do I need to be technical to use it?
You do not need to be technical to use the assistant once it is set up and you are messaging it from a familiar chat app. Installing it, choosing a model, connecting accounts, and tightening permissions is technical work. That is the part we do.
Can OpenClaw send email or manage a calendar?
The official site describes an assistant that can organize an inbox, send emails, and manage a calendar, among other jobs, when it is connected to those tools. Whether it should do that in your business depends on the accounts you connect and the approval rules you set. We recommend starting with drafts or internal work before anything goes to a customer on its own.
How is OpenClaw different from ChatGPT?
ChatGPT is a hosted assistant you use in an app or a browser. OpenClaw is software you run on your machine and message from chat apps you already use. It can take actions on that computer, keep memory and skills locally, and work with models you bring. They solve different problems, and one is not a replacement for the other in every case.
Do you only set up OpenClaw?
No. We review what you already use and recommend what fits. We have also worked with Hermes Agent and Grok Bot and set those up for clients. If a well-known assistant is the better starting point, we will say so.