This weekDots became available by default in eligible ChatGPT Enterprise, Edu and Healthcare workspaces unless an admin opts out.Loblaw became the first Canadian retailer to connect to Muse, and voice reached Muse users in Canada.Changelog

How to install OpenClaw

OpenClaw is a free, MIT-licensed AI assistant from the OpenClaw Foundation that you install on your own computer or server, and anyone can download it. You talk to it through chat apps you already use, such as WhatsApp, Telegram, Slack and iMessage, and it can run tools and shell commands on the machine it lives on. The software costs nothing, but you pay for whichever model and tool APIs you connect, unless you run a local model.

Steps summarised from the official docs, checked October 20, 2026. Commands change between releases, so check the docs link before you paste anything.

  1. Check requirementsInstallers cover macOS, Linux, WSL2 and Windows PowerShell. If you install through npm instead, you need Node.js 24.16+ or 26.1+, with Node 26 recommended.
  2. InstallOn macOS, Linux or WSL2 run: curl -fsSL https://openclaw.ai/install.sh | bash. On Windows PowerShell run: iwr -useb https://openclaw.ai/install.ps1 | iex. The script detects your OS, adds Node if missing and starts onboarding.
  3. Finish onboardingPick Quick start, which reuses detected AI access and tests it with a completion, or Custom setup for the full guided flow. Run openclaw onboard --classic for the older step-by-step wizard.
  4. Run the gateway as a serviceQuick start leaves the gateway in the foreground. Press Ctrl+C, then run openclaw gateway install to keep it running in the background.
  5. Open the dashboardRun openclaw dashboard to open the Control UI in your browser and send a test message.
  6. Connect TelegramFollow the docs' Telegram setup page to create the bot and add its token. Telegram uses long polling by default, and its DM policy defaults to pairing.
  7. Approve your first DMRun openclaw pairing list telegram, then openclaw pairing approve telegram <CODE>. Codes are 8 characters and expire after 1 hour, or approve under Settings > Channels > DM access requests in the Control UI.
  8. Lock it downRun openclaw security audit, keep the gateway on loopback, configure a sandbox such as Docker or Podman before giving it shell work, and read any ClawHub skill before installing it.

Before you connect it to anything

Where it runs
On your own machine or server; the gateway binds to loopback on a normal host install, while container images default to an exposed bind and need auth.
What it can reach
Tools run on the host for the main session unless you configure a sandbox (Docker, Podman, SSH and others are documented).
When it asks you
Unknown DM senders get a pairing code you must approve (openclaw pairing approve); groups are allowlisted and usually mention-gated. Exec approvals are documented on a separate page we did not read.

Official sources

  1. The official repository is openclaw/openclaw on GitHub and it is MIT-licensed. github.com read Oct 20, 2026
  2. The README says the project is developed by the OpenClaw Foundation, described as an independent 501(c)(3), with donors including Amazon, OpenAI, Red Hat and others. github.com read Oct 20, 2026
  3. The docs list mainstream channels Discord, Google Chat, iMessage, Matrix, Microsoft Teams, Signal, Slack, SMS, Telegram, WhatsApp and X, plus many self-hosted and regional ones (LINE, WeChat, Feishu, Zalo and more). docs.openclaw.ai read Oct 20, 2026
  4. Installers cover macOS, Linux, WSL2 and Windows PowerShell; npm installs need Node.js 24.16+ or 26.1+, with Node 26 recommended. github.com read Oct 20, 2026
  5. The docs describe costs as coming from paid provider APIs (models, media, embeddings, web search, speech, third-party skills) rather than from OpenClaw itself. docs.openclaw.ai read Oct 20, 2026
  6. The latest stable GitHub release is openclaw 2026.11.5, published 17 November 2026. github.com read Nov 19, 2026
  7. Unknown senders on DM-capable channels are paired by default and must be approved by the owner. github.com read Oct 20, 2026
  8. Tools run directly on the host for the main session unless sandboxing is configured. github.com read Oct 20, 2026
  9. The security overview says agents with the message tool can send across conversations and providers by default unless restricted. docs.openclaw.ai read Oct 20, 2026
  10. The GitHub security advisories page lists roughly 640+ published advisories (65 pages); the newest shown, dated 30 June 2026, include several rated High for authorization bypasses. github.com read Oct 20, 2026