Beginner's Guide to OpenClaw: Getting Started in 2026

How This Guide Was Built

This guide answers the question “how do I get started with OpenClaw?” step by step. It is based on official OpenClaw documentation, the project’s GitHub repository, and community reports from August 2026. We did not run the tool hands-on; every fact links to a primary source so you can verify it yourself.

Sources: official docs (docs.openclaw.ai), the GitHub repository, the MIT license file, and the credits page.
Verified: every external link below returned HTTP 200 when checked on August 13, 2026.
Not tested: we did not install or run OpenClaw ourselves; the steps come from the official quickstart.
Freshness: facts reflect the current release as of August 2026; Node version floors and install commands were re-checked against the official docs on 2026-08-13.

What OpenClaw Does and When to Use It

Here’s the one-sentence pitch: OpenClaw is a personal AI assistant you run on your own computer, and you talk to it from your favorite chat apps like Telegram, Discord, or WhatsApp. It runs a lightweight local gateway that connects those apps to the AI model of your choice, so your messages, files, and automations stay on hardware you control.

Concretely, OpenClaw runs a lightweight server — the “gateway” — on your machine at port 18789 (Source). You connect your chat app to that gateway, and the gateway connects to an AI model of your choice (like OpenAI, Anthropic, or a local model). You send a message in Telegram, OpenClaw forwards it to the AI, and the reply comes back to your chat. Simple.

The big selling point is owning your data. Because OpenClaw runs on your hardware, your conversation history, your files, and your automations stay with you. Nothing is stored on a third-party cloud server unless you choose a cloud-hosted model (Source).

Who is this for? OpenClaw is for people who are comfortable with a terminal, want privacy, and enjoy tinkering. It’s also for people who want to automate tasks — scheduling messages, monitoring feeds, or building simple workflows. It is not for absolute beginners who have never opened a command line.

The maintainers are blunt about this. Their official stance: “if you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely” (Source). That’s not gatekeeping — it’s a warning. OpenClaw can execute commands on your machine. If you don’t know what you’re doing, you can break things.

Use OpenClaw when you want a self-hosted assistant with real privacy, deep customization, and the ability to connect to many chat platforms. If you just want a quick chatbot in your browser, use a hosted service instead. For a deeper look, check our OpenClaw review.

What OpenClaw Costs

The software itself is free — OpenClaw is released under the MIT license (Source), which means you can use, modify, and distribute it without paying a cent. There are no paid tiers, no premium features, and no subscription required (Source). Your only real costs are the AI tokens you consume and, optionally, a VPS if you want it online 24/7.

So where do costs come from? Two places:

LLM tokens. OpenClaw doesn’t include a brain. You bring your own AI model API key (OpenAI, Anthropic, etc.), and you pay the provider per token. If you use a local model via Ollama, this cost drops to near zero — you’re just paying for electricity (Source).
Optional hosting. If you want OpenClaw running 24/7 without keeping your computer on, you’ll rent a VPS (virtual private server). That typically costs $5–$20 per month depending on the provider. We have a guide to install OpenClaw on a VPS in 10 minutes if you go that route.

The project is supported by sponsors including OpenAI, GitHub, NVIDIA, and Vercel, plus community donations (Source). This funding keeps the core free. In short: you can get started for $0 upfront, and your ongoing cost is whatever you spend on AI tokens.

How Do I Get Started with OpenClaw?

This is the section you came for. Follow these seven steps in order, and you’ll have OpenClaw talking to you in Telegram within about five minutes (Source). You’ll need a terminal, an API key from a model provider, and a Telegram account. If you hit a snag, each step notes the most common failure and how to fix it.

Step 1: Check Your Node.js Version

The #1 beginner failure is installing OpenClaw with an outdated Node.js. OpenClaw requires Node.js 22.22.3 or newer, 24.15 or newer, or 25.9 or newer. The recommended version is Node 26 (Source).

To check your version, open a terminal and run:

node --version

If you see something like v20.11.0, you’re too old. If you see v22.22.3 or higher, you’re fine. If not, go to nodejs.org and install the latest LTS. Don’t skip this step — it’s the most common reason installs fail.

Step 2: Get an API Key

OpenClaw doesn’t include a built-in AI. You need to bring your own key from a provider like OpenAI, Anthropic, or Google. Sign up for the provider of your choice, generate an API key, and keep it handy. You’ll paste it in during onboarding.

This key is what makes OpenClaw’s “live completion” tests work during setup. The onboarding wizard will send a test request to your chosen model to verify everything is connected (Source). If you don’t have a key, the wizard can’t validate your setup.

Step 3: Install OpenClaw

Installation depends on your operating system.

macOS or Linux: Open a terminal and run:

curl -fsSL https://openclaw.ai/install.sh | bash

Windows: Open PowerShell as Administrator and run:

iwr -useb https://openclaw.ai/install.ps1 | iex

Alternative (any OS with Node): Use npm directly:

npm install -g openclaw@latest

Important: The package name is openclaw, not @openclaw/openclaw. The older scoped name is deprecated and will not work. Use openclaw@latest to get the current release (Source).

Step 4: Onboard

With OpenClaw installed, run the onboarding command:

openclaw onboard --install-daemon

The --install-daemon flag sets OpenClaw to run in the background automatically, so you don’t have to start it manually every time (Source).

The wizard will ask you several questions:

Which AI provider and model to use
Your API key
Which chat channels to connect (you can add more later)
Whether to enable the daemon

Answer each prompt. The wizard tests your API key with a live completion to confirm it works (Source).

Step 5: Verify the Gateway

After onboarding, confirm the gateway is running:

openclaw gateway status

You should see output that includes Runtime: running. If you see stopped or error, check your Node version and API key first. The gateway is the heart of OpenClaw — if it’s not running, nothing else works (Source).

Step 6: Send Your First Message

Open the built-in dashboard:

openclaw dashboard

This opens your browser at http://127.0.0.1:18789/. The dashboard gives you a chat interface to talk to your AI directly.

Send a simple message like “Hello” or “What can you do?” and confirm you get a reply. Get one clean conversation working here before you connect Telegram. If something is broken, it’s much easier to debug in the dashboard than through a chat app. Once the dashboard works, move on to Telegram.

Step 7: Connect Telegram

Now for the fun part — talking to OpenClaw from your phone.

Open Telegram and search for @BotFather (the official bot that creates other bots).
Send /newbot and follow the prompts to name your bot. You’ll get a token — a long string of characters. Save it.
In your OpenClaw dashboard, navigate to the Telegram channel settings.
Paste your bot token into the configuration. OpenClaw will create a pairing request.
Open your new bot in Telegram and send it /start. Then approve the pairing request in the dashboard.

The pairing code expires after one hour, so don’t wait too long between steps (Source).

Once paired, you can message your bot from anywhere and OpenClaw will respond using your configured AI model. That’s it — you’re live.

OpenClaw vs Hermes Agent, Kimi, and Symphony

How does OpenClaw compare to other tools in the space? If you’re choosing between a self-hosted agent, a hosted assistant, and a dev-team orchestrator, the differences matter more than the marketing. Here’s a quick comparison table, followed by deeper analysis of each pairing and who each tool actually fits.

	OpenClaw	Hermes Agent	Kimi	Symphony
What it is	Self-hosted AI agent gateway; connects LLMs to 20+ messaging channels	Open-source agent with learning loop (skills + memory + cron)	Hosted AI assistant + open-weight models (K2.5/K3)	Open-source orchestration framework for Codex coding runs
License & cost	MIT, free; pay only LLM tokens	MIT, free; pay only LLM tokens	Free tier + paid API; closed hosted product	Open-source; runs Codex so OpenAI API costs apply
Beginner fit	Needs terminal + API key; 5-min install; huge community	Needs terminal; 5-min install; hermes setup —portal	Easiest — nothing to install, browser/app only	Hardest — dev-team tool; no chat channels
Messaging channels	20+ (Telegram, Discord, WhatsApp, Signal, Slack, iMessage)	Telegram/Discord/Slack + 20+ via gateway	Built into its own app; no external gateway	None (Linear/GitHub tickets in, PRs out)
Setup time	~5 minutes to first Telegram message	~5 minutes to first chat	0 minutes (sign up in browser)	Hours — project/team/infra setup required
Notables	386k GitHub stars; OpenClaw Foundation; iOS/Android apps	Learning loop, Nous Portal; blog has Hermes guide	Usable inside OpenClaw as model provider	Codex-based; not a personal assistant

OpenClaw vs Hermes Agent: These two are the closest competitors. Both are MIT-licensed, self-hosted, and free. Hermes Agent’s standout feature is its learning loop — it stores skills and memories to improve over time (Source). OpenClaw’s edge is its massive community (386k+ GitHub stars (Source)) and broader channel support. For beginners, Hermes might be slightly easier because of its hermes setup --portal command, but OpenClaw has more tutorials and community help. Read our OpenClaw vs Hermes Agent comparison for a deeper dive, or start with our Hermes Agent beginner’s guide to compare both setups side by side.

OpenClaw vs Kimi: Kimi is a hosted product — you use it in a browser or app, and there’s nothing to install. That makes it the easiest option for beginners. But you don’t own your data, and you can’t connect it to Telegram or other chat apps. Interestingly, Kimi’s models can be used inside OpenClaw as a provider, so they’re not mutually exclusive.

OpenClaw vs Symphony: Symphony is a completely different beast. It’s an orchestration framework for running Codex (OpenAI’s coding agent) at scale. It’s designed for development teams, not personal assistants. It has no chat channels — you create tickets in Linear or GitHub, and Symphony produces pull requests. Setup takes hours. If you’re a solo beginner, this is not for you.

Common Beginner Mistakes

Here are the seven most common mistakes beginners make, based on community reports and official troubleshooting guides (Source). Each row pairs the mistake with its symptom so you can recognize it fast, and a concrete fix you can apply in under a minute. Bookmark this table; most of these errors are one-line repairs.

Mistake	Symptom	Fix
Wrong Node version	Install fails with cryptic errors	Check `node --version`; upgrade to Node 22.22.3+ or 26 (Source)
Missing API key	Onboarding hangs at “test completion”	Generate a key from your LLM provider before starting
Skipping daemon install	OpenClaw stops when you close the terminal	Re-run `openclaw onboard --install-daemon`
Port conflict	Gateway fails to start; “port 18789 in use”	Find and kill the process using port 18789, or change the port in config
Malformed JSON	Config file errors on startup	Validate your `openclaw.json` with a JSON linter
Discord 401/403	Bot can’t connect to Discord	Enable Message Content Intent in Discord developer portal (Source)
Expecting zero cost	Surprised by API bills	Remember: OpenClaw is free, but LLM tokens aren’t; use local models via Ollama to cut costs

Pro Tips for Your First Week

Once you’re up and running, here’s how to get the most out of OpenClaw. These five tips cover the settings, habits, and add-ons that save beginners the most time — from picking the right update channel to cutting your token bill with local models.

Don’t bind to LAN without auth. By default, OpenClaw listens on localhost. If you change it to listen on your network, you must set up authentication first. See the security section below.

Use update channels. OpenClaw has two update channels: stable and dev. For daily use, stick with stable:

openclaw update --channel stable

Dev has newer features but more bugs (Source).

Try local models with Ollama. If you want to save money on API tokens, install Ollama and run a local model. OpenClaw can connect to it, and your cost drops to near zero. It’s slower than cloud models, but it’s free and fully private.

Install companion apps. OpenClaw has official apps for macOS, Windows, iOS, and Android (Source). The mobile apps are especially useful — you can check your assistant’s status and send messages from your phone without opening Telegram.

Read the automation guide. OpenClaw’s cron and heartbeat features let you schedule tasks. Check out our cron and heartbeat automation guide for ideas.

Security: Lock It Down Before You Walk Away

OpenClaw has powerful capabilities, and with power comes responsibility. It can run shell commands, read and write files, and send messages as you, so locking down access is step one — not an afterthought. Here’s what you need to know before you walk away from your setup.

Pairing vs allowlist. OpenClaw uses a dmPolicy system to control who can talk to your bot. The default is “pairing” — users must request access and you approve them. The alternative is an “allowlist” — only pre-approved users can connect. For most beginners, pairing is safer because it requires explicit approval (Source).

Non-loopback binds require auth. If you configure OpenClaw to listen on a network interface other than localhost (e.g., to access it from your phone on the same Wi-Fi), you must enable authentication. Without it, anyone on your network could control your assistant. The official docs are explicit about this requirement (Source).

Vet third-party skills. OpenClaw supports “skills” — plugins that extend its capabilities. Some skills execute code or access files. Before installing a third-party skill, review its source code. Cisco’s AI security research team demonstrated that a malicious skill could exfiltrate data and inject prompts without the user noticing, precisely because skill repositories lacked adequate vetting (Source). Don’t install random skills from strangers.

Remember the maintainer’s warning: “if you can’t understand how to run a command line, this is far too dangerous of a project for you to use safely” (Source). OpenClaw can run shell commands. It can read and write files. It can send messages as you. Treat it like a powerful tool — respect it, and it’ll serve you well.

Keep your API key secret. Your key is stored in the config file. Don’t commit it to GitHub, don’t share it, and rotate it if you suspect a leak.

FAQ

These are the questions beginners ask most often, answered directly. If you’re deciding whether to install OpenClaw, worried about cost, or unsure how safe it is to run on your own machine, start here. Each answer links to the official documentation for the full details.

Do I need an account to use OpenClaw?

No. OpenClaw is self-hosted software, so there’s no account, signup, or cloud lock-in. You do need two things: a computer that can run Node.js, and an API key from an LLM provider such as OpenAI or Anthropic, which bills you separately. The onboarding wizard walks you through both (Source).

How much does it cost?

The software is free under the MIT license, with no paid tiers or subscriptions (Source). You pay only for the LLM tokens your provider bills, so a light setup costs a few dollars a month, and a local model via Ollama costs nearly nothing. Optional VPS hosting for 24/7 uptime runs about $5–$20 per month (Source).

Is OpenClaw safe to run on my computer?

OpenClaw is open-source and widely audited, but it can execute shell commands, read and write files, and send messages as you. Safety comes from configuration: keep the default pairing policy, require authentication before binding to anything but localhost, and vet third-party skills before installing them (Source). Follow the security section above and the risk is manageable.

Conclusion

OpenClaw is a powerful, free, and private way to run your own AI assistant. The setup takes about five minutes, and once it’s running, the command line stops feeling intimidating — you’re in control of your own AI. For the full picture, read What Is OpenClaw?; to compare options, see the 10 best OpenClaw alternatives.

NiteAgent — AI agent development, frameworks, and production patterns
Hermes Tutorials — Hermes Agent setup, configuration, and advanced workflows
NoCode Insider — AI workflow automation with no-code tools, agents, and APIs

Cross-links automatically generated from None.

Back to all posts