1300 166 138 Tell us what you need
Resources / OpenClaw

Meet OpenClaw: The Open Source AI Agent You Can Run Locally

OpenClaw open source AI agent framework running locally

🎯 QUICK PICK

• OpenClaw: open source AI agent framework, runs 100% locally
• Created by Peter Steinberger, MIT licence, free software plus your own model API key
• Connects WhatsApp, Telegram, Slack, Microsoft Teams, Signal, Discord, iMessage and more
• Skills system for extending capabilities via ClawHub marketplace
• Requires Node 26, or Node 22.22.3+, 24.15+ or 25.9+; setup takes 15-20 minutes

On this page

OpenClaw is an open source framework that puts an AI agent inside the messaging apps your team already uses, from WhatsApp and Telegram to Slack, Microsoft Teams and Signal, while the agent itself runs on your own hardware. For Australian businesses weighing the Privacy Act against the convenience of cloud assistants, that local first design is the main reason to look at it closely.

I look at tools like this from one angle first: where the data goes. OpenClaw answers that better than most, but it is not a polished product. It needs a computer that stays on, some comfort with the command line and an API key from a model provider, and it carries real security risks if you connect it to the wrong accounts.

This guide covers how OpenClaw works, how it compares with Manus, AutoGPT, Open Interpreter and Claude Computer Use, how to install it, what its skills system adds, where it fits for Australian businesses and how to start without getting overwhelmed. The frequently asked questions at the end cover cost, hardware, safety, the messaging apps it supports and who is behind the project.

What is OpenClaw and how does it work?

OpenClaw is an open source framework that connects AI agents to your everyday communication channels, from WhatsApp, Telegram, Slack and Microsoft Teams to Signal, Discord and iMessage, while keeping everything running on your own hardware. Created by Peter Steinberger, the framework acts as a gateway between messaging platforms and a local AI agent that can read files, execute code, browse the web, and automate tasks on your behalf.

The architecture is straightforward. A single Gateway process runs on your machine and keeps the connections to your messaging accounts. When a message arrives, the Gateway routes it to an embedded AI agent, powered by the runtime bundled with OpenClaw (older configuration files still call it pi, now a deprecated alias). The agent uses whatever tools it needs and replies through the same channel.

What sets OpenClaw apart from cloud based assistants is the local first design: there is no external server in the middle of the core agent logic, and your conversations, files and credentials stay on your machine. The agent works inside a workspace directory that holds its context, memories and configuration. For secrets it should never see in plain config, use a dedicated password manager.

  • Skills: modular instruction sets stored as markdown files that teach the agent how to use specific tools or follow particular workflows. You can install community skills from ClawHub, the project’s public registry, or write your own, so the framework is highly customisable without changes to the core codebase.
  • Multi agent setups: different agents can handle different channels or tasks, each with isolated workspaces and credentials, and a single Gateway can host several side by side, routing inbound messages through configurable bindings.
  • Governance: since July 2026 the project has been stewarded by the OpenClaw Foundation, an independent nonprofit with no paid tier and no hosted service.
Comparison of OpenClaw with other AI agent frameworks

OpenClaw vs other AI agent frameworks

The AI agent landscape has exploded over the past two years, and choosing the right framework depends on what you actually need it to do. The short version: OpenClaw is the one built around your messaging apps and your own hardware. Here is how it compares with the four alternatives people most often weigh it against, and when each one is the better fit.

OpenClaw vs Manus AI

  • Where it runs: Manus made headlines as a fully autonomous agent that completes complex tasks with minimal human input, but it runs as a cloud service, while OpenClaw runs entirely on your hardware.
  • Choose Manus for a more polished out of the box experience, accepting that your data goes to external servers with every interaction. Choose OpenClaw if data sovereignty matters, as it does for many Australian businesses.

OpenClaw vs AutoGPT

  • What it is: AutoGPT was one of the first autonomous agent frameworks to gain mainstream attention. Both projects are open source, but AutoGPT focuses on goal driven autonomous loops: you give it an objective and it works out the steps.
  • Choose OpenClaw if you want an agent that responds to your WhatsApp messages and executes tasks on demand, with a human in the loop. Choose AutoGPT if you want to set a goal and walk away.

OpenClaw vs Open Interpreter

  • What it is: Open Interpreter lets you run code locally through a chat interface, similar to ChatGPT’s Code Interpreter but on your own machine.
  • The difference: OpenClaw integrates with coding agents like Codex, GitHub Copilot CLI and Claude Code and can do similar things, but adds the messaging layer on top. Think of Open Interpreter as a local coding assistant and OpenClaw as a full agent gateway that can include coding alongside other tools.

OpenClaw vs Claude Computer Use

  • What it is: Anthropic’s Computer Use feature allows Claude to control a computer directly, clicking, typing and navigating applications.
  • The difference: OpenClaw relies on APIs, command line tools and direct file access rather than visual automation. Computer Use is impressive for tasks inside graphical interfaces, but slower and more error prone than direct tool access, so for most automation OpenClaw’s approach is faster and more reliable.

The bottom line

OpenClaw is not trying to be the most autonomous or the most capable agent framework. It is designed for practical, daily use. An AI assistant that lives in your messaging apps, respects your privacy, and executes tasks when you ask. If that matches your needs, it is hard to beat.

Installing OpenClaw on a local computer

How to install OpenClaw

Setting up OpenClaw is simpler than most open source AI tools, though it does require some comfort with basic computer administration. The framework runs on macOS, Linux and Windows, where you can start with the native Windows Hub app, the PowerShell installer or a WSL2 Gateway, and there are companion apps for iOS and Android.

What you need before starting

  • A supported Node.js release: OpenClaw now requires Node 24.16 or later, or Node 26.1 or later, and recommends Node 26, which starts the Gateway faster and uses less memory. The Node 22 and Node 25 builds it used to accept are no longer supported. If Node is missing, the installer provisions Node 26 on macOS and the Node 24 LTS line on Linux. Node itself is free.
  • A computer that stays on, or a small server.
  • Accounts for whichever messaging platforms you want to connect.
  • An API key from a model provider such as Anthropic, OpenAI or Google. The provider bills that usage separately.

The setup process

Installation happens through the terminal, but OpenClaw includes a guided wizard that walks you through each step. The wizard creates your configuration files, sets up the workspace where the agent stores its memory and instructions, and installs a background service so everything runs automatically when your computer starts.

Connecting your messaging apps

  • WhatsApp connects the same way you would set up WhatsApp Web: you scan a QR code with your phone and the link is established.
  • Telegram and Discord require creating a bot through their respective platforms, which takes about five minutes each.
  • iMessage works on Mac only and needs a small additional tool, imsg, which OpenClaw’s official iMessage plugin uses.

Once it is running

After setup, OpenClaw runs quietly in the background, with a browser based dashboard to monitor conversations, adjust settings and check that everything is working. The real interaction happens through your normal messaging apps. For most users the initial setup takes about fifteen to twenty minutes, and our OpenClaw documentation has step by step guides with screenshots for each platform.

OpenClaw skills system extending AI agent capabilities

OpenClaw skills and capabilities

Out of the box, OpenClaw can read and write files, execute commands, browse the web, and handle media like images and voice notes. But the real power comes from Skills, modular addons written in plain markdown that tell the agent how to use a particular tool or follow a specific workflow.

  • Want your agent to check the weather, manage your calendar or call a specific API? There is usually a skill for that, or you can write your own.
  • Each skill lives in a simple folder with a configuration file that defines when the agent should use it.
  • Skills can include scripts, reference documents and detailed instructions: everything the agent needs to handle a particular task competently.

What are Skills?

ClawHub: the skills marketplace

  • ClawHub is a public registry where developers share skills for free. You can browse them at clawhub.com and install them with a single command.
  • Popular skills cover common use cases like weather forecasts, web searching, file management and integrations with productivity tools.
  • Installing a skill takes seconds, you can update all your installed skills at once when new versions are released, and if you build something useful you can publish it for others.

Built in capabilities

Even without additional skills, OpenClaw agents can read documents and summarise them, write and edit text, search through files on your computer, run calculations and hold context across long conversations, remembering what you shared earlier in a session. For Australian businesses that means an agent that drafts emails in your voice, prepares meeting notes, organises files and handles routine queries without sending your data to external servers.

Customising behaviour

  • SOUL.md defines the agent’s personality and communication style.
  • AGENTS.md provides operating instructions and persistent memory.
  • USER.md stores information about you so the agent can personalise its responses.
  • These workspace files are plain text and easy to edit. No programming required.

Security considerations

While OpenClaw keeps your data local, it is not immune to security risks. One significant concern is prompt injection through external content. If the agent has access to your email, an attacker could send a message containing hidden instructions, text formatted to be invisible to humans but readable by the AI, and the agent might follow them without realising they came from a malicious source.

My recommendation: do not connect OpenClaw to email accounts, banking systems or any service containing sensitive credentials. Limit the agent’s access to what it genuinely needs, and be cautious about giving it permission to act on external content it has not verified.
Australian businesses using OpenClaw for local AI automation

Using OpenClaw for Australian businesses

For Australian organisations the conversation around AI tools always comes back to one question: where does the data go? With OpenClaw, nowhere: everything stays on your hardware. Australian businesses operating under the Privacy Act, handling health records or working on government contracts face strict rules on storage and processing, and cloud assistants send queries, documents and conversations through overseas servers.

Data sovereignty and compliance

OpenClaw sidesteps this entirely. The agent runs locally, processes information locally and stores everything locally, so if you work with sensitive client information such as legal documents, financial records or medical data, you can use AI assistance without creating new compliance risks. For teams that will stay on ChatGPT, masking client details before they are sent is the lighter alternative.

Practical use cases

  • A financial adviser in Melbourne uses OpenClaw to draft client correspondence and summarise meeting notes.
  • A recruitment agency in Sydney has its agent screen CVs and schedule interviews through WhatsApp.
  • A logistics company in Brisbane automates its daily reporting by having the agent pull data from spreadsheets and format updates.
  • The common thread: small and medium businesses saving hours each week on information they cannot afford to leak, without compromising client confidentiality.

Integration with Australian tools

OpenClaw works alongside the software Australian businesses already use. It can connect with Xero for accounting queries, integrate with local CRM systems, and handle Australian date and time formats correctly. Because skills are customisable, you can build workflows specific to your industry or adapt existing skills to work with Australian platforms.

Cost considerations

  • Running OpenClaw requires a computer that stays on: a desktop, a small server, or a cloud instance you control. For most small businesses an old computer or a low cost local server is enough.
  • The software itself is free under an MIT licence; you pay only your model provider for API usage.
  • Compared with per seat subscriptions for enterprise AI tools, the total cost of ownership is often significantly lower.
Getting started with OpenClaw AI agent setup

Getting started with OpenClaw

If OpenClaw sounds like a fit for your needs, here is the path I recommend for getting started without getting overwhelmed.

  1. Start with the documentation. The official OpenClaw documentation at docs.openclaw.ai covers everything from basic setup to advanced configurations. Begin with the Getting Started guide, which walks through installation step by step and includes troubleshooting for common issues.
  2. Join the community. OpenClaw has an active Discord community where developers and users share tips, answer questions and showcase what they have built. If you hit a roadblock during setup or want advice on a particular use case, it is responsive and welcoming to newcomers.
  3. Begin with one channel. Rather than connecting every messaging platform at once, start with the one you use most. Get comfortable with how the agent responds, experiment with basic tasks and learn how the workspace files affect behaviour. Adding more channels afterwards is straightforward.
  4. Explore skills gradually. Browse ClawHub, but resist the urge to install everything. Start with one or two skills that address your immediate needs and expand from there: too many at once make it harder to understand what the agent is doing and why.
  5. Customise your workspace. Spend time editing SOUL.md and AGENTS.md to shape how your agent communicates. This is where OpenClaw becomes genuinely useful: an assistant that matches your preferences and context. The default settings work, but the personalised version works better.
  6. Consider your hardware. For personal use, your main computer is fine. For business use, dedicate a small server or always on machine so the agent stays available when your laptop is closed. A basic setup does not need powerful hardware.

Local AI agents are just getting started

OpenClaw represents a shift: instead of trading privacy for capability, organisations can run AI agents entirely on their own infrastructure. It is not for everyone, as it needs technical comfort and lacks commercial polish, but for teams handling sensitive data those tradeoffs often make sense. Hosted options have matured too, and the commercial AI agents you can use in Australia now include Gemini Spark, ChatGPT Dots and Claude Cowork.

Australian businesses exploring AI automation should consider where their data flows before choosing a platform, and OpenClaw offers one answer worth evaluating. For organisations that want guidance on implementing AI agents, whether OpenClaw or another solution, Aivy helps Australian businesses navigate these decisions with practical, privacy conscious advice.

Ready to explore what AI agents could do for your business? Book a free AI Discovery Session with our team. For weekly insights on AI tools and automation, subscribe to our newsletter.

Frequently asked questions about OpenClaw

What is OpenClaw?

OpenClaw is an open source AI agent framework that connects messaging platforms like WhatsApp, Telegram, and Discord to a locally running AI assistant. All data stays on your own hardware.

Is OpenClaw free to use?

The software is free. OpenClaw is released under the MIT licence, so you can download, use, and modify it at no cost. Running it also needs an API key from a model provider such as Anthropic, OpenAI or Google, and that provider bills the usage separately.

Who created OpenClaw?

OpenClaw was created by Peter Steinberger, a well known developer in the Apple and open source communities. Since July 2026 the project has been stewarded by the OpenClaw Foundation, an independent nonprofit, and Steinberger still leads its technical decisions.

Does OpenClaw require coding skills?

Basic setup requires comfort with command line tools but no programming. Customising the agent through workspace files uses plain text. Building custom skills may require some coding knowledge.

What messaging apps work with OpenClaw?

OpenClaw supports WhatsApp, Telegram, Slack, Microsoft Teams, Signal, Discord, Google Chat, Matrix, iMessage and Zalo. Further channels, including Mattermost, Twitch and Nostr, are available as plugins.

Is OpenClaw safe for business use?

OpenClaw runs locally, so data stays on your infrastructure. However, connecting it to email or sensitive accounts carries risks. Attackers can embed hidden instructions in emails (invisible to humans but readable by the agent), potentially tricking it into sharing confidential information. For business use, limit what the agent can access and avoid connecting it to email, banking, or systems containing credentials.

How does OpenClaw compare to ChatGPT?

ChatGPT runs in the cloud and processes data on external servers. OpenClaw runs locally, giving you full control over your data. ChatGPT offers a polished interface; OpenClaw offers privacy and customisation.

Can OpenClaw control my computer?

Yes. OpenClaw agents can read and write files, execute commands, browse the web, and automate tasks on your machine, all within the permissions you configure.

What hardware do I need for OpenClaw?

Any modern computer running macOS, Linux or Windows (natively or through WSL2) with Node 24.16 or later, or Node 26.1 or later; Node 26 is recommended, and the installer adds it if Node is missing. For always on availability, a dedicated small server or repurposed desktop works well.

Can multiple people use the same OpenClaw agent?

Yes. OpenClaw supports multi agent setups where different users or teams have isolated agents, or a shared agent can respond to multiple authorised users.

What is ClawHub?

ClawHub is the public registry for OpenClaw skills. Developers share free skills that add functionality, and users can install them with a single command.

Last updated: 19 October 2026

On this page

Most Read

NEWSLETTER

The Aivy Brief

AI news and practical guides for Australian businesses. No hype, no spam.

Free. Unsubscribe anytime.

Scroll to Top