There are three ways to run OpenClaw. This guide covers the simplest. I went from zero to a working agent connected to Telegram in 25 minutes using Hostinger’s one-click setup, and I have written every step here in order.
Before You Start: What You Need
No matter which path you choose, have these ready before you begin:
- An API key from an AI provider (Anthropic, OpenAI, or Google), or a Nexos AI Credits account if you plan to use Hostinger’s integrated billing
- A Telegram account for the channel setup covered in this guide
- About 30 minutes on your first attempt, 10 minutes on any repeat deployment
If you are going with a VPS or one-click install, you will also need a credit card for the hosting cost.
3 Ways to Set Up Your OpenClaw AI Agent
1. Local Hardware
Running OpenClaw on your own machine means complete control over your data, no hosting bill, and the option to run a local AI model via Ollama instead of sending any data to external APIs.
What you need:
- Docker installed (Docker Desktop works on Mac and Windows; Docker Engine on Linux)
- An AI API key, or Ollama set up with a compatible local model
- The machine staying powered on whenever you need the agent reachable
To install, pull the OpenClaw Docker image and follow the official documentation. The initial image pull takes a few minutes; after that, OpenClaw starts in under a minute.
When this makes sense:
- You are testing OpenClaw before committing to paid hosting
- Data privacy is the top priority and nothing can leave your hardware
- You are building a private document assistant using local models
When it does not work:
- You need scheduled tasks that run on a fixed time each day
- You want to message your agent from your phone while you are away from your desk
- The agent needs to monitor anything while your machine is off or asleep
2. Virtual Private Server
A VPS gives you an always-on server in the cloud. You install Docker yourself, configure environment variables, connect your API keys, and manage everything over SSH. You pick the provider, the region, and the specs.
OpenClaw runs on 1 vCPU and 2GB RAM at minimum, though 4GB gives comfortable headroom once you add skills and scheduled tasks.
| Provider | Monthly price | RAM | OpenClaw guide |
| Hostinger KVM 1 | ~$6 | 4 GB | Official + one-click option |
| Contabo VPS | ~$5–6 | 8 GB | Community scripts |
| Vultr Regular | ~$12 | 2 GB | Community |
| DigitalOcean Droplet | ~$32 | 2–4 GB | Official + Marketplace image |
| OVHcloud VPS | ~$6–10 | 2–4 GB | Community |
A few practical distinctions worth knowing:
- Best price-to-RAM ratio: Contabo gives you 8GB RAM for $5–6/month, which is unusual value for the spec. No official template, but community Docker scripts make setup manageable.
- Best for EU users on a budget: Hetzner plans have an official OpenClaw guide and consistently low pricing. Limited to EU and US data centers.
- Best global coverage: Vultr has one of the widest regional footprints if you need a server where other providers do not operate.
- Best documentation: DigitalOcean costs more, but their Linux and Docker guides are thorough. Worth it if you expect to troubleshoot.
- Hostinger: Strong value, official guide, and the only provider here that also offers a dedicated one-click OpenClaw template, so you can start with Option 3 and switch to manual control later without changing providers.
3: One-Click Hosting
One-click hosting handles Docker installation, environment configuration, and container startup through a web interface. No terminal required.
Hostinger currently offers the most complete OpenClaw-specific deployment experience. Their dedicated OpenClaw product page walks through plan selection, gateway token retrieval, and API key configuration in a single flow.
DigitalOcean offers OpenClaw through their Marketplace as a one-click Droplet image. You get full root access and strong documentation. The main trade-off is cost: a comparable Droplet costs three to four times as much per month as Hostinger.
Bluehost markets a VPS with one-click OpenClaw deployment, including NVMe storage and unmetered bandwidth. A practical option if you are already using Bluehost for other hosting and want to keep everything under one dashboard.
Contabo has no official one-click option, but community-maintained scripts produce a similar result once you are SSH’d in. Best suited to users who want the cheapest possible infrastructure and are comfortable with a terminal.
Fully managed options (no server management at all): xCloud handles setup, updates, and security automatically. clawoneclick.com and openclawd.ai offer hosted OpenClaw as a service, often bundled with higher-level automation workflows. Higher monthly cost than self-hosted, but zero infrastructure overhead.
Tip: If you already have a Hostinger VPS from another project, you do not need to buy a new server. Go to hPanel, open Docker Manager, click OS & Panel, then Change OS, search for OpenClaw, and deploy from there.
How to Set Up OpenClaw with Hostinger’s 1-Click Hosting
What follows is the full walkthrough, from selecting a plan to messaging your agent from Telegram.
Step 1: Choose Your Plan
Go to Hostinger’s OpenClaw page, and you will see two products:
Managed OpenClaw. Fully managed hosting. Hostinger handles setup, security, updates, and backups. You get built-in Telegram and WhatsApp pairing, pre-installed AI credits, and a dedicated agentic email inbox. No terminal access, no configuration. Good if you want OpenClaw running with zero technical work.
OpenClaw on VPS. A self-managed VPS with OpenClaw pre-installed via template. You get full root and terminal access, 2 vCPU, 8GB RAM, 100GB NVMe storage, and 8TB bandwidth. Good if you want full control over your environment, the ability to install custom skills, or plan to run other services alongside OpenClaw.
This guide follows the OpenClaw on VPS path. Click Choose plan on that option.
Step 2: Pick Your Add-Ons and Server Location
On the next page, three things need your attention.
Billing period: Twelve or twenty-four months gives you the lowest monthly rate. The one-month plan is available if you want to test before committing, but the per-month savings on longer terms are significant.
Daily backups: Enable this. OpenClaw lets you reconfigure its server environment through conversation, which means it can occasionally make changes that break the setup. A daily backup gives you a reliable rollback point. If you skip it during checkout, you can add it later from hPanel for a small monthly fee.
Server location: Pick the region closest to you or your primary audience. Hostinger shows estimated latency for each option, which makes the decision easy. For a personal agent where only you are the user, choose the closest region. Germany, Singapore, and US East Coast tend to have the widest range of surrounding services if you plan to add other tools later.
Step 3: Complete Account Setup and Payment
Create your Hostinger account or log in to an existing one, enter your billing address, and complete payment. From first click to confirmed order typically takes under two minutes.
Step 4: Save Your Gateway Token
Once payment confirms, Hostinger redirects you to the OpenClaw configuration screen inside hPanel. The first thing you see is your OpenClaw Gateway Token.
This token is the only key to your agent’s dashboard. Anyone who has it can access your agent.
- Click the eye icon to reveal it
- Copy it before doing anything else
- Save it in a password manager
After deployment, the gateway token is accessible from the VPS Overview page in hPanel. The Gateway token button copies it to your clipboard instantly. You will not need to dig through menus.
If you ever need to find it manually: hPanel > VPS > Docker Manager > Projects > your OpenClaw project > Environment tab > eye icon next to OPENCLAW_GATEWAY_TOKEN.
Step 5: Connect an AI Model
Still on the configuration screen, you need to connect an AI model. Two options.
Option A: Nexos AI Credits (zero extra setup)
If you add Nexos AI Credits at checkout, they are automatically wired into your deployment. OpenClaw connects to AI providers through Nexos, which handles authentication on your behalf. Google’s Gemini Flash is the default model, with access to Claude and others through the same credits.
The cost model is usage-based. Based on my testing, a session of 92 messages covering web searches, code generation, file operations, and system monitoring cost $0.45 total.
At that average rate, you would need around 4,000 messages per month to reach $20, which is what a single standalone AI subscription costs.
Nexos makes sense if:
- You want the fastest possible setup with zero separate accounts to manage
- You are running light to medium usage, well under a thousand messages per day
- You prefer one billing source for both the server and the AI usage
Option B: Your Own API Key
If you prefer direct access to a provider, paste your key into the corresponding field on the configuration screen.
For Anthropic (Claude):
- Go to console.anthropic.com and create an account
- Under Manage > Billing, add credits. Start with at least $40 to reach Tier 2 rate limits, which are noticeably more permissive during a setup session
- Under Manage > API keys, generate a key and copy it immediately
- Paste it into the Anthropic field on the Hostinger config screen
For OpenAI (GPT-4o):
- Go to platform.openai.com and create an account
- Add billing credits under Settings > Billing
- Generate a key under API keys
- Paste it into the OpenAI field
For Google (Gemini):
- Go to aistudio.google.com
- Generate an API key from the API Keys section
- Paste it into the Google field
You can add multiple provider keys during setup and switch between models inside the OpenClaw dashboard later.
Tip: If you are connecting only one provider, Claude Sonnet handles reasoning and writing tasks well. GPT-4o is strong across the board, including code. Gemini Flash is the most cost-efficient for high-volume scheduled tasks where speed matters more than depth.
Step 6: Deploy and Wait
Click Deploy.
A progress indicator appears while Hostinger provisions the server, installs Docker, pulls the OpenClaw image, and starts the container. The full process takes three to four minutes. When the status turns green and shows Running, the server is live.
Step 7: Open Your Dashboard
From the hPanel VPS Overview page, you have two shortcuts next to your OpenClaw project:
- OpenClaw: opens your dashboard login in a new tab
- Gateway token: copies your token to the clipboard with one click
Click OpenClaw.
The WebSocket URL is pre-filled. Paste your gateway token into the Gateway Token field and click Connect.
You land on the Chat tab. Before sending anything, click Overview in the sidebar and confirm the STATUS shows OK in green. That confirms the dashboard is communicating with your gateway correctly.
The sidebar gives you access to everything:
- Chat: direct conversation with your agent
- Overview: gateway health, uptime, skill count, recent sessions
- Instances: which agents are running
- Sessions: full conversation history across all channels
- Usage: token consumption and cost broken down by session and model
- Cron Jobs: scheduled task management
- Agents: AI model and behavior configuration
- Skills: installed capabilities
- Nodes: node management
- Docs: in-app documentation
Step 8: Verify Your Model Connection
Go to Agents in the sidebar. If you used Nexos AI Credits, you will see a pre-configured agent named “main” with a model already selected and active. The integration is working; nothing more to do.
If you added your own API key during configuration, confirm it appears correctly. The model dropdown will show available models for that provider. If the key was not picked up, you can paste it directly in the Agents section without redeploying.
Step 9: Connect Your Channels and APIs
This is where OpenClaw changes from a web dashboard into something you can reach from anywhere.
The most direct way to connect a messaging channel in the current version of OpenClaw is conversational. You describe what you want to connect, and the agent walks you through it.
In the Chat tab, type:
I want to connect to Telegram
The agent returns a numbered sequence. Here is what you do:
In Telegram:
- Open Telegram and search for @BotFather. Look for the blue verification checkmark.
- Send /newbot
- Choose a display name for your bot. This is what shows up in the chat header.
- Choose a username. It must end in bot, for example, myagent_bot. Telegram requires it to be unique.
- BotFather returns a bot token. Copy it and treat it like a password. Anyone who has it can control your bot.
Back in OpenClaw Chat: Paste the token into the chat and send it.
OpenClaw runs the connection commands in the background. Within seconds, it confirms the connection is live.
Back in Telegram: Search for your bot by its username, tap Start, and send any message. The bot responds with a pairing code.
Copy the pairing code. Go back to OpenClaw’s Chat tab and paste it. The agent completes the pairing.
Now test it. Go to Telegram and send a message. Your agent responds.
Connecting other channels:
Using the same conversational approach, tell your agent which platform you want and it walks you through the steps:
- Discord: create a bot application in Discord’s Developer Portal, enable the three privileged gateway intents (message content, server members, presence), then paste the bot token into the OpenClaw chat. One thing to check after connecting: Discord’s Group Policy in OpenClaw defaults to “allowlist,” which means the bot receives messages but responds to no one until you add users or channels to the allowed list. Go to Channels > Discord > Group Policy and change it to “open” if you want the bot to respond to all server members.
- WhatsApp: link via QR code scan, similar to WhatsApp Web
- Slack: create a Slack app, configure OAuth scopes and bot permissions, paste the bot token
- Signal and iMessage (Mac only) also have integration paths through the same conversational setup flow
Step 10: Install Skills As You Need Them
OpenClaw ships with a meaningful set of skills pre-installed. Based on the capability list the agent returns when you ask what it can do on a fresh server, here is what works without any extra setup:
Ready immediately, no credentials:
- Web search and web fetch for live information retrieval
- Weather lookups via wttr.in
- Memos, diagrams, PDF generation, and document summarization
- Python and Node.js code debugging and editing
Available but need accounts or API keys:
- Messaging: Discord, Slack, Telegram, WhatsApp, iMessage, Apple Notes, Obsidian, Notion, Trello, GitHub Issues
- Voice and media: ElevenLabs text-to-speech, OpenAI Whisper transcription, Spotify, Sonos, Hue smart lights
- Server health auditing: SSH hardening checks, firewall review, exposure scanning
Finding and installing more skills:
Browse the official marketplace at clawhub.ai to see all available skills with descriptions and security information. Or ask your agent directly: What skills would be useful for someone managing a content team?
To install a skill, ask by name: Install the Google Calendar skill. The agent walks you through any credentials or configuration it needs.
Before installing anything from community publishers, check the VirusTotal security report on the skill’s ClawHub page. Your agent will often prompt you to do this before proceeding. Do not skip it.
Step 11: Secure Your Agent Before It Does Real Work
An agent with file access, API keys, and a server connection can cause damage if misconfigured. Take twenty minutes on this before handing it anything meaningful.
Limit to what you actually need. If you are using OpenClaw as a personal assistant for scheduling and research, it does not need shell access. Disable capabilities you are not using.
Set channel access controls. If your bot is in a shared Discord server or Telegram group, configure the allowlist. An open bot that responds to everyone in a shared space is a problem waiting to surface.
Keep API keys out of conversation history. If the Agents section asks for a key, add it there. Never paste credentials into the chat.
Keep logging on. OpenClaw logs agent actions by default. This lets you audit what ran, when it ran, and what it produced. If something unexpected happens, the logs show you exactly what the agent did.
Create a snapshot before experimenting. In hPanel, go to VPS > Backups & Monitoring > Snapshot section and create one before you install new skills or make configuration changes. If something breaks, you restore to that point in seconds instead of rebuilding from scratch.
How to Turn OpenClaw Into Your 24/7 AI Agent
At this point, OpenClaw is running and you can message it from Telegram. That makes it a chat tool. What makes it a 24/7 agent is different: instead of you starting every conversation, the agent acts on its own, on a schedule, and messages you when something needs your attention.
This section is about setting that up.
Understand the shift: you stop initiating, the agent does
Most people use OpenClaw like a smarter search engine at first. They open Telegram, ask a question, get an answer. That is useful, but it is not what a 24/7 agent actually is.
A 24/7 agent works in three modes simultaneously:
Scheduled: It does something at a set time without you asking. Your morning brief arrives at 7am. A weekly metrics report posts to Slack every Monday. A content ideas list lands in your inbox every Sunday. You never ask. It just happens.
Triggered: It acts when a condition is met. Disk usage hits 85 percent, it messages you. A GitHub build fails, it sends you the error and a link. A high-priority email arrives after hours, it flags it. You find out because it told you, not because you went looking.
On-demand: You ask, it handles it. This is what most people start with, and it stays useful. But the first two modes are what make it a 24/7 agent rather than a chat tool.
Getting there means telling the agent what to do when you are not watching.
Set up your first scheduled automation
The fastest way to create a scheduled task is to tell your agent what you want in plain language. In the Telegram chat or the OpenClaw web dashboard, send something like:
Every weekday at 7:00am, send me a Telegram message with today’s weather in [city], my first three calendar events, and the top three headlines from [news source]. Keep it under 150 words.
The agent confirms the schedule, sets up the cron job, and starts running it at the next trigger. You do not need to write a cron expression or touch any configuration file.
To verify it is running, open the Cron Jobs tab in the OpenClaw dashboard. Each job shows its schedule, last run time, and whether it succeeded.
The Overview tab shows a count of active jobs alongside your gateway status. If STATUS shows OK in green, the agent is live and your scheduled tasks will fire.
If a scheduled task does not fire when expected, check two things first: whether the Cron Jobs tab shows it as active, and whether there is a timezone mismatch between what you told the agent during onboarding and what the server clock is running on. You can check the server timezone by asking your agent directly: What timezone is this server set to?
What works well as a 24/7 automation
Not every task is worth scheduling. The ones that work best share a common trait: they produce something useful on a fixed rhythm, or they catch something you would not notice until it was already a problem.
Daily and weekly outputs:
- Every morning at 7:00am, send me a brief with today’s weather, my first three calendar events, and top headlines from [source]. Under 150 words.
- Every Monday at 9:00am, post last week’s key metrics from [dashboard] to our #updates Slack channel with week-on-week change for each number.
- Every Friday at 5:00pm, summarize what I accomplished this week based on our conversation history and any tasks I marked done.
Monitoring and alerts:
- Monitor this VPS every 30 minutes. If CPU stays above 80 percent for two consecutive checks, disk passes 85 percent, or RAM exceeds 90 percent, send me a Telegram alert immediately.
- Watch my GitHub Actions workflows. If any build fails or a production deploy completes, notify me on Telegram with the commit message and a link to the run.
- Check my inbox every morning at 8:00am. If anything marked urgent arrived since yesterday, summarize it and send me a Telegram message.
Recurring research and content:
- Every Sunday at 6:00pm, give me five article ideas for [topic]. For each one, write a headline and two sentences on why the timing is right.
- On the first of each month, search for the current pricing and key changes for [tool A], [tool B], and [tool C] and send me a comparison.
Hostinger’s OpenClaw use cases page covers 25 more worked examples across productivity, developer workflows, and business automation.
Give the agent what it needs to act
Scheduled tasks only work if the agent has access to the things it needs. A morning brief that includes your calendar requires the Google Calendar skill to be installed and authorized. A weekly metrics report requires a connection to wherever those metrics live.
Before you set up an automation, check whether the relevant skill is installed by asking your agent: Do you have access to [Google Calendar / my email / GitHub]? If not, it will tell you what to install and walk you through the authorization.
For tasks that pull from dashboards or internal tools without a direct integration, OpenClaw can use browser automation or web fetch to retrieve the data.
Ask your agent to describe how it would approach the task before you set it up as a scheduled job. That conversation often surfaces what is missing before the first failed run.
Start With One Automation
The setup is done. You have OpenClaw running on Hostinger, connected to Telegram, and ready for its first task.
The temptation is to configure everything at once. Resist it. Pick one scheduled automation, run it for a week, and see what you actually find useful before adding more. A morning brief, a server health alert, a weekly report. Something that runs without you and delivers something real. Build from there.
The agent compounds in value over time, not all at once.
