What is OpenClaw?

OpenClaw is one of the newer tools in the class of AI agents: software that does not only answer questions but takes actions. This page describes it using the project’s own repository, as of November 2026.

What it is

The OpenClaw repository describes it as an open-source AI assistant that runs on your own computer and meets you in the channels you already use. Rather than living in a browser tab, it works through those channels. The project lists Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ more, plus native apps for macOS, iOS, Android, Windows, and Linux.

At the center is a local Gateway, which the project describes as the local control plane for sessions, tools, events, and channel connections. You can run it as a personal assistant on a laptop or set it up for a team.

Who runs it

The project says it is stewarded by the OpenClaw Foundation, an independent 501(c)(3), and has no paid tier, hosted service, or token. You (or your IT team) run it yourself.

Data and updates

The OpenClaw README says your prompts go to the model provider and chat platforms you configure. The OpenClaw README says by default OpenClaw itself phones home for nothing but a daily version check. It also says models and agent harnesses such as Claude, Codex and local models are plugins you can swap.

What people use an assistant like this for

With an agent that lives in your messaging apps and can call tools, common uses include:

• Asking for a summary of a document by sending it in a chat.
• Having recurring tasks run and report back in Slack or a messaging channel.
• Reviewing documents such as contracts. In our class, AI tools are used for a first pass on provided sample agreements, flagging clauses against a provided checklist; see whether AI can review contracts.

The exact capabilities depend on which tools and AI models you connect, so read the documentation for your version.

Security: read this before installing

The OpenClaw security docs say that on a regular host install the Gateway binds to loopback. The OpenClaw security docs say most chat channels answer an unknown DM sender with a pairing code instead of processing the message. The OpenClaw security docs say group access is allowlisted, usually behind a mention gate.

The repository is direct about risk, and you should be too:

• The OpenClaw README says to treat inbound messages as untrusted input, including direct messages from unknown senders.
• Direct-message channels pair unknown senders by default, and you approve each pairing request.
• The OpenClaw README says tools run on the host for the main session unless you configure sandboxing.
• Review the security guide, exposure runbook and sandboxing guide before connecting other users or exposing the Gateway remotely.

In plain terms: a tool that can act on your computer and accepts messages from outside is a place where mistakes have consequences. Do not connect it to work accounts or data without approval, and start with a machine or account that has nothing sensitive on it.

How it fits with other tools

Claude Code and Codex are agent tools built around working on files in a folder; OpenClaw is oriented around a personal assistant reachable through chat channels. See what AI agents are and the OpenClaw vs Claude Code comparison.

Should you use it?

If you are comfortable with some technical setup and you want an assistant you control, it is worth exploring. If you want something your employer has already vetted, start with tools your IT team supports.

OpenClaw is one of the tools that may be mentioned as an example in Learn AI Fast; the day-two document review lab does not depend on it.

Sources

• GitHub, openclaw/openclaw README: “OpenClaw is an open-source AI assistant that runs on your own computer and meets you in the channels you already use”
• GitHub, openclaw/openclaw README: “Discord, iMessage, Slack, Teams, Telegram, WhatsApp, and 20+ more, plus native apps for macOS, iOS, Android, Windows, and Linux”
• GitHub, openclaw/openclaw README: “is the local control plane for sessions, tools, events, and channel connections”
• GitHub, openclaw/openclaw README: “an independent 501(c)(3), and has no paid tier, hosted service, or token”
• GitHub, openclaw/openclaw README: “Your prompts go to the model provider and chat platforms you configure”
• GitHub, openclaw/openclaw README: “by default OpenClaw itself phones home for nothing but a daily version check”
• GitHub, openclaw/openclaw README: “Models and agent harnesses (Claude, Codex, local models) are plugins you can swap”
• OpenClaw docs, Security: “On a regular host install the Gateway binds to loopback”
• OpenClaw docs, Security: “most chat channels answer an unknown DM sender with a pairing code instead of processing the message”
• OpenClaw docs, Security: “group access is allowlisted, usually behind a mention gate”
• GitHub, openclaw/openclaw README: “Treat inbound messages as untrusted input”
• GitHub, openclaw/openclaw README: “DM-capable channels pair unknown senders by default”
• GitHub, openclaw/openclaw README: “Tools run on the host for the main session unless you configure sandboxing”