Most AI chat apps can talk about clearing your inbox or checking a server, but OpenClaw actually does it from the chat app you already use. So, what is OpenClaw? OpenClaw is a free, open-source personal AI agent that runs on your own machine or server and takes real actions for you through apps like Telegram, WhatsApp and Slack.
It grew fast. The project launched in late 2025 and passed 390,000 GitHub stars by November 2026. If you are already choosing between agents, our Hermes Agent vs OpenClaw comparison goes deeper on that decision.
This guide is the plain explainer. You will learn where OpenClaw came from, how its gateway, channels, skills and memory fit together, what people use it for and how it compares with other agents. Then you will see what it takes to keep it running 24/7.
TL;DR: What Is OpenClaw in 60 Seconds
Short on time? Here is the whole guide in a few lines:
- OpenClaw is an open-source AI agent (MIT license) that lives on your hardware and acts for you, instead of only answering questions.
- One Gateway process connects your chat apps, your AI model and the agent’s tools, such as commands, files and a browser.
- You talk to it where you already chat: Telegram, WhatsApp, Slack, Discord, Signal, iMessage, Microsoft Teams and many more.
- Skills add new abilities. Each skill is a small folder with instructions, and ClawHub is the public registry where people share them.
- Memory is plain Markdown files you can read, edit and back up yourself.
- It is free, but not free to run. You pay for an AI model and an always-on server, and your time is the hidden cost.
What Is OpenClaw?
OpenClaw is a self-hosted AI assistant that connects a large language model to real tools and to the messaging apps you use every day. Its own tagline is “The AI that really does things.” You bring the model (Claude, GPT, Gemini or a local one), and OpenClaw gives it hands.
Think of it like hiring a personal assistant who works from your office. ChatGPT is a smart advisor you visit in a browser tab. OpenClaw is the assistant at the desk next to yours: it can open the filing cabinet, use the computer and text you when something needs your attention.
A few things worth knowing:
- Austrian developer Peter Steinberger created it. He is best known for founding PSPDFKit, a PDF toolkit company.
- It has had three names. It launched as Clawdbot in November 2025, became Moltbot in January 2026 and settled on OpenClaw a few days later. Reports say the first rename followed a naming concern from Anthropic, since “Clawd” sounded close to “Claude.”
- It now has a foundation. In February 2026, Steinberger announced he was joining OpenAI and that OpenClaw would move to an independent open-source foundation. The project stays under the MIT license.
- The mascot is a space lobster named Molty, which explains the claw puns everywhere.
OpenClaw at a Glance
| Detail | Value |
|---|---|
| Type | Open-source, self-hosted personal AI agent |
| Created by | Peter Steinberger |
| First release | November 2025, as Clawdbot |
| License | MIT |
| Built with | TypeScript, running on Node.js |
| Source code | openclaw/openclaw on GitHub |
| GitHub stars | About 394,000 (as of November 1, 2026) |
| Stewardship | Independent open-source foundation (announced February 2026) |
| Memory format | Plain Markdown files in the agent’s workspace |
| Minimum RAM on xCloud | 4 GB |
How OpenClaw Works: Gateway, Channels, Skills and Memory
OpenClaw looks like a chatbot from the outside. Inside, it is four parts working together.
The Gateway
The Gateway is the heart of OpenClaw. Its README calls it the local control plane that ties sessions, tools, events and channel connections together. In practice, it is one long-running process that receives your messages, passes them to the model and runs whatever tools the model asks for.
You manage the Gateway through a web Control UI, a command line or a terminal interface. Because it is one process, it needs a machine that stays on. Close your laptop and the agent goes quiet.
Channels
Channels are the chat apps OpenClaw listens on. The project lists WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Microsoft Teams and Google Chat, plus 20+ more. Companion apps for macOS, iOS, Android, Windows and Linux add voice, a live Canvas and device actions.
Here is the part that trips people up. Unknown senders on direct-message channels must be approved first through a pairing step. That default stops a stranger from finding your bot and giving it orders.
Skills and ClawHub
A skill is a folder with a SKILL.md file that tells the agent how and when to use a tool. OpenClaw follows the open AgentSkills format, so skills are readable text, not hidden code. Some skills ship with the install, and you can write your own.
ClawHub is the public registry where the community shares skills. You install one with openclaw skills install @owner/skill-name or simply ask the agent in chat. Our guide to OpenClaw skills and ClawHub on xCloud shows the full flow.
Memory
OpenClaw remembers things by writing plain Markdown files in its workspace. MEMORY.md holds long-term facts and decisions, and dated daily notes hold the running detail. A search tool mixes meaning-based and keyword matching to find old notes when they matter.
The catch worth knowing: the model only knows what was saved to disk. If an agent seems forgetful, it usually never wrote the fact down. Our post on why your OpenClaw agent keeps forgetting covers the fix.
6 OpenClaw Features That Matter
1. Real actions, not just answers
OpenClaw can run shell commands, read and write files, browse the web and call APIs. That is exactly what makes it powerful, and risky if deployed carelessly. Tools run directly on the host unless you turn on sandboxing.
2. Any model you like
OpenClaw works with hosted and local model providers. Models such as Claude, Codex and local models plug in and can be swapped without rebuilding your setup. You pay your provider directly, with no markup from OpenClaw.
3. Proactive automations
OpenClaw can work without being asked. On first setup, it creates a “Proactive check” automation that runs every 30 minutes, and you can add your own scheduled jobs. Good first jobs include a daily briefing, an uptime check or a weekly report.
4. Memory you can open and edit
Because memory is Markdown, you can read it, fix a wrong fact or delete a topic. You can also back it up like any other folder.
5. One agent, many chat apps
The same agent answers you on Telegram at your desk and on WhatsApp on the train. You do not set up a separate bot per platform. This breadth is one of OpenClaw’s clearest strengths.
6. An open, extendable core
Beyond skills, developers can build plugins with the plugin SDK and share them through ClawHub. Everything is MIT licensed, so you can read, change and reuse the code.
What Is OpenClaw Used For?
Personal admin
People use OpenClaw to sort and summarize email, handle calendar invites, set reminders and pull key points out of long documents. The value is that it happens from a chat message, not a new app.
Research and daily briefings
Ask it to watch a topic, check a few sources each morning and message you a short summary. Scheduled jobs make this a set-and-forget routine.
Developer and DevOps work
Developers point OpenClaw at code tasks, log checks and server health reports. With the xCloud skill from ClawHub, it can even manage your hosting from chat, so you can ask for a site backup or a server check in plain words.
Small business automation
Founders and agencies use it for lead follow-ups, report drafts and team updates in Slack. For 15 worked examples, including a real SaaS app the agent helped build, read our OpenClaw use cases guide.
OpenClaw vs Hermes Agent vs ChatGPT
OpenClaw is not the only agent worth a look. Here is a fair snapshot of how it sits next to two common alternatives:
| Factor | OpenClaw | Hermes Agent | ChatGPT (app) |
|---|---|---|---|
| What it is | Self-hosted personal AI agent | Self-hosted AI agent from Nous Research | Hosted AI assistant |
| License | MIT, open source | MIT, open source | Proprietary |
| Where it runs | Your machine or server | Your machine or server | OpenAI’s cloud |
| Main strength | Many chat channels, ClawHub skills | Built-in learning loop that writes its own skills | Zero setup, polished apps |
| Memory | Markdown files you control | Agent-curated memory plus session search | Managed by OpenAI |
| Models | Any hosted or local model | Many providers, including local | OpenAI models |
| Needs a server? | Yes, always on | Yes, always on | No |
If you only remember one line: OpenClaw optimizes for reach and control, Hermes Agent for learning over time and ChatGPT for convenience. None is universally better. To go deeper on the second option, see what Hermes Agent is.
What You Need to Self-Host OpenClaw
OpenClaw itself is light, because the model usually runs at your provider. Still, a dependable setup needs a few things:
- An always-on machine. A spare computer works for testing. A VPS works better for 24/7 use.
- Enough memory for tools and browsing. xCloud requires at least 4 GB RAM for OpenClaw servers.
- A recent Node.js. The README asked for Node 24 or newer as of November 2026.
- A model provider key, such as Anthropic, OpenAI or Gemini, or a local model.
- A chat app bot, for example a Telegram bot token.
Ways to run it
The DIY route is one line. Run curl -fsSL https://openclaw.ai/install.sh | bash, then openclaw onboard --install-daemon to set it up as a background service. That part takes minutes.
The real work starts after setup. You still need a domain, SSL, firewall rules, updates, a safe pairing setup and a plan for when the Gateway crashes.
Your time is the hidden cost, and the occasional 3 a.m. restart is part of the bill.
The managed route hands that work to a host. For a side-by-side of cost and effort, read managed vs self-hosting OpenClaw. For a ranked list of hosts, see the best OpenClaw hosting providers.
Stay safe either way
OpenClaw can act with real permissions on its server. Use separate accounts for testing, not your main email or GitHub. Our OpenClaw security best practices guide lists seven steps worth doing on day one.
OpenClaw Pros and Cons
| Pros | Cons |
|---|---|
| ✅ Free and open source under MIT | ❌ Needs an always-on machine to be useful |
| ✅ Works in the chat apps you already use | ❌ Real system access means real security risk |
| ✅ Any hosted or local model | ❌ Model costs depend on how much it works |
| ✅ Readable Markdown memory you control | ❌ Memory only holds what the agent saves |
| ✅ Large community and ClawHub skill registry | ❌ Community skills need review before you trust them |
| ✅ Proactive scheduled automations | ❌ Self-hosting means updates and fixes are on you |
Launch Your OpenClaw Agent on xCloud in About 5 Minutes
If you want OpenClaw without the server work, OpenClaw hosting on xCloud gives the agent its own dedicated server. xCloud handles the server setup, the OpenClaw install, SSL and updates. Managed plans start at $9.99/month for the first purchase of a 6 GB Cloud VPS (renews at $19.99/month), and you bring your own AI provider key.
Here is how the one-click deploy works:
- Start a new OpenClaw server from the OpenClaw checkout and give it a name.
- Choose a plan under Server Size (4 GB RAM or more) and an in-stock Region.
- Confirm the security warning after reading it.
- Pick your AI Provider, such as Anthropic, Claude Code, OpenAI, OpenRouter, Moonshot AI (Kimi) or Google Gemini, and paste your key.
- Add your Telegram Bot Token and Telegram User ID (optional). Your account is paired for you, with no commands to run.
- Choose whether the Control UI dashboard is reachable, then click Create.
When it finishes, the Status page shows your dashboard URL and login details. From the same dashboard you can change Providers and Channels, edit Environment settings, run Updates, read Logs and reset OpenClaw if a setup breaks.
A few honest limits apply. The server runs only OpenClaw, Telegram is the built-in chat channel today and our support covers the server, not your prompts or workflows.
Prefer your own VPS? xCloud can also install OpenClaw on a server you already own, without Docker.
Ready to chat with your own agent today? Deploy OpenClaw on xCloud and skip the server setup.
👉 Follow the step-by-step guide to deploy OpenClaw with xCloud.
Put Your First OpenClaw Agent to Work This Week
OpenClaw is the open-source agent that turned “an AI that does things” into something you can run yourself. Its Gateway connects your chat apps, your model and real tools. Skills add abilities, and Markdown memory keeps context you can see.
The decision is less about the software and more about where it lives. Self-host it if you enjoy the setup and want full control. Use a managed host if you want the agent answering on Telegram by lunch, not by the weekend.
Whichever path you pick, start small. Give it one routine task this week, such as a morning briefing, and grow from there.
If you have found this blog helpful, feel free to subscribe to our blogs for valuable tutorials, guides, knowledge and tips on web hosting and server management. You can also join our Facebook community to share insights and engage in discussions.
Frequently asked questions
What is OpenClaw in simple words?
OpenClaw is a free, open-source AI assistant that runs on your own computer or server. You chat with it in apps like Telegram or WhatsApp. It can do real work for you, such as running commands, managing files and checking the web on a schedule.
Is OpenClaw free?
Yes. The software is free and open source under the MIT license. You still pay for two things: an AI model provider (or your own local model) and a computer or server that stays online so the agent can answer you.
Is OpenClaw the same as Clawdbot and Moltbot?
Yes. The project launched as Clawdbot in November 2025, was briefly renamed Moltbot in January 2026 and settled on the name OpenClaw a few days later. The code, the community and the lobster mascot carried over through both renames.
Does OpenClaw need a powerful computer?
No, because the heavy AI work usually happens at your model provider. The agent itself needs an always-on machine with enough memory for its gateway, tools and browser tasks. xCloud runs OpenClaw on servers with at least 4 GB RAM.
Is OpenClaw safe to use?
It is as safe as your setup. OpenClaw can run commands and edit files, so treat every incoming message as untrusted, keep the pairing approval on and use separate accounts instead of your personal ones. Our OpenClaw security best practices guide walks through seven concrete steps.
What is the difference between OpenClaw and Hermes Agent?
Both are open-source, self-hosted AI agents under the MIT license. OpenClaw focuses on reaching you on many chat apps and lets you install skills from ClawHub. Hermes Agent, from Nous Research, focuses on a built-in learning loop that writes new skills from its own experience.
I want OpenClaw running on Telegram without touching a terminal. What should I do?
Use a managed OpenClaw host. On xCloud, you pick a server size, add your AI provider key, paste your Telegram bot token and user ID, then click Create. The agent is usually ready in about 5 minutes.