How to Deploy OpenClaw — Step by Step Guide

OpenClaw is an open-source personal AI agent that runs on your own machine. It can connect to Telegram, WhatsApp, Discord, and more — acting as your 24/7 AI assistant that can read files, browse the web, control your browser, and automate tasks.

Think of it as having your own ChatGPT that lives on your computer, has access to your files, and can take real actions on your behalf.

This is the most important step — and most beginners skip it.

Before you even install OpenClaw, you should have your AI model API keys ready. OpenClaw needs an API key to connect to an AI model (like Claude, GPT-4, Qwen, etc.) — and getting these keys takes time. If you try to set up OpenClaw without one, you'll be stuck at the model selection step.

Why prepare first?

• Some platforms require phone verification, real-name registration, or credit card binding — this can take hours or even days.
• Free tiers have strict token limits. Without a proper plan, your agent may stop working after a few messages.
• Getting everything ready beforehand means your first OpenClaw experience is smooth — install → configure → chat, all in one sitting.

Below are the recommended subscription plans for first-time users. We recommend starting with a monthly subscription plan so you can control costs and avoid surprise token bills.

Here are the best options for first-time OpenClaw users, ranked by ease of setup:

Option 1: OpenAI — API Key

• Platform: OpenAI Platform
• Plan: Pay-as-you-go (minimum top-up US$5 / ~S$6.70) or ChatGPT Plus (US$20/month / ~S$26.80 for web + API access)
• Models: GPT-5.5 "Spud" (flagship, April 2026), GPT-5.4 mini (fast & cheap), o3/o4-mini (deep reasoning), plus legacy GPT-4o series
• Why it's great: GPT-5.5 is OpenAI's most advanced model — excels at complex reasoning, coding, and multi-modal tasks (text + images). The o-series reasoning models are best for math, science, and step-by-step problem solving. Most widely supported across all OpenClaw features.
• Setup time: ~10 minutes (account + credit card)

Option 2: Claude (Anthropic) — API Key

• Platform: Anthropic Console
• Plan: Pay-as-you-go API or Claude Pro (US$20/month / ~S$26.80 for web + bonus API credits)
• Models: Claude Sonnet 4.6 (best value, US$3/MTok input), Claude Opus 4.7 (most powerful, US$5/MTok input), Claude Haiku 4.5 (fastest & cheapest, US$1/MTok input)
• Why it's great: Claude Sonnet 4.6 is the sweet spot — excellent reasoning and coding at a reasonable price. Opus 4.7 is the most capable model for complex tasks. Claude excels at long document analysis, agent workflows, and structured output. Built-in prompt caching reduces costs by up to 90% for repeated content.
• Setup time: ~15 minutes (account + billing setup)

Option 3: OpenRouter — One Key, 300+ Models

• Platform: OpenRouter
• Plan: Pay-as-you-go, no minimum. Top up from US$10 / ~S$13.40. Transparent passthrough pricing (provider cost + small markup).
• Models: 300+ models including GPT-5.5, Claude Opus 4.7 / Sonnet 4.6, Gemini 2.5, Qwen, Llama 4, Mistral, DeepSeek, and dozens more open-source models.
• Why it's great: One API key gives you access to every major model provider. Perfect for testing and comparing models side-by-side without managing multiple accounts. Auto-routing picks the best available provider for each model. Ideal if you want flexibility to switch between Claude, GPT, and open-source models.
• Setup time: ~5 minutes (account + top-up)

Other Great Options

• Alibaba Cloud Bailian (阿里云百炼): Qwen models with competitive pay-as-you-go pricing. Note: The Coding Plan (monthly subscription) has been discontinued — now uses pay-as-you-go billing. bailian.console.aliyun.com
• Google Gemini API: Gemini 2.5 with generous free tier (15 RPM, 1M tokens/day). AI Studio
• DeepSeek API: DeepSeek V4 released April 2026. V4-Flash at just US$0.14/MTok input, V4-Pro at US$1.74/MTok (75% off until May 31). Both support 1M context and OpenAI/Anthropic API formats. Extremely competitive pricing with strong Chinese language support. platform.deepseek.com
• SiliconFlow (硅基流动): Free tier with multiple model options. cloud.siliconflow.cn

💡 Pro tip: For your first deployment, we recommend OpenAI or Claude (easiest setup, best docs) or OpenRouter (one key for all models). OpenRouter is especially useful for testing different models before committing to a specific provider.

Before you start, make sure you have:

Open your terminal and run:

npm install -g openclaw

sudo npm install -g openclaw

npm install -g openclaw

This installs OpenClaw globally on your system. Wait for the installation to complete — it may take 1-2 minutes.

Verify the installation:

openclaw --version

You should see the version number printed.

Run the setup wizard:

openclaw setup

This will guide you through:

• Choosing your AI model provider (OpenAI, Claude, Gemini, Qwen, etc.)
• Entering your API key (the one you prepared earlier!)
• Setting your workspace directory

For Telegram:

1. Message @BotFather on Telegram
2. Send /newbot and follow the prompts
3. Copy the bot token (looks like 123456:ABC-DEF...)
4. Paste it when OpenClaw asks for it

For the model API key:

• If using OpenAI: Go to OpenAI API Keys → Create new secret key
• If using Claude: Go to Anthropic Console → Create API Key
• If using OpenRouter: Go to OpenRouter Keys → Create Key
• If using Google Gemini: Go to AI Studio API Keys → Create API Key
• If using DeepSeek: Go to DeepSeek API Keys → Create Key

Start the OpenClaw gateway:

openclaw gateway start

You should see output like:

✓ Gateway started on port 3000
✓ Connected to Telegram
✓ Agent ready — send a message to your bot!

Now message your Telegram bot — it should respond!

Send a message to your bot on Telegram. Try:

• Hello! — basic greeting test
• What's the weather in Singapore? — web search test
• Search for "best hawker centres SG" — browsing test

If you get responses, congratulations — OpenClaw is deployed! 🎉

Skills extend what your AI can do. Install them with:

openclaw skills install <skill-name>

Popular skills to start with:

• browser-automation — control web browsers
• github — interact with GitHub
• pdf — read and process PDF files
• weather — weather forecasts
• slack — connect to Slack

See our next article for the Top 10 Must-Install Skills.

Since you're using a subscription plan, here's how to keep costs under control:

• Set usage alerts: Most platforms let you set spending limits. Do this immediately after creating your API key.
• Use cheaper models for routine tasks: For example, use GPT-5.4 mini or Claude Haiku 4.5 for everyday chat, and save Claude Sonnet 4.6 or GPT-5.5 for complex reasoning.
• Monitor your token usage: Check your platform's dashboard regularly. OpenClaw logs token usage too — use openclaw logs to review.
• Start with free tiers: Gemini API has a generous free tier. Claude Haiku is very affordable for testing. Test with free/cheap models first, then upgrade.
• Use OpenRouter for model comparison: One key to try 300+ models. Compare pricing and quality before committing to a single provider.

Bot not responding?

• Check the gateway is running: openclaw gateway status
• Verify your bot token is correct
• Check logs: openclaw logs

Model errors or "API key invalid"?

• Run openclaw models list to see available models
• Set a model: openclaw models set chatgpt
• Verify your API key is active and has quota remaining
• Check if your API credits have run out (OpenRouter, pay-as-you-go providers)

Running out of tokens too fast?

• Switch to a cheaper model (e.g., GPT-5.4 mini, Claude Haiku 4.5, Qwen via OpenRouter)
• Check your usage dashboard and set limits
• Consider topping up or trying a different provider on OpenRouter

Need more help? Join the OpenClaw Discord community.