Self-hosting guide
How to self-host OpenClaw on Hetzner with ClawKickstart
Deploy OpenClaw on your own Hetzner VPS with ClawKickstart: connect your project, choose a server, configure your model and Telegram, then verify a working agent.
You can run OpenClaw on a Hetzner VPS in your own cloud account and use Telegram to talk to it. This guide follows the ClawKickstart dashboard: connect Hetzner, provision the server, configure your model and messaging credentials, then hatch the agent and verify its first reply. You do not need a domain, DNS records or a public gateway for this setup.
ClawKickstart is a paid guided setup service. The one-time setup purchase, your Hetzner server bill and model-provider usage are separate costs. If you prefer to install everything manually, use the official OpenClaw Hetzner deployment guide linked below; its commands and maintenance steps follow a different path from our dashboard.
1. Prepare your accounts and credentials
- A ClawKickstart account and a setup purchase for the instance you want to deploy.
- A Hetzner Cloud account and project, with permission to create cloud resources.
- A model-provider account, an API key and the exact model reference you want to use. Check that billing or credits are enabled and that the key can access that model.
- For Telegram, a bot token from BotFather. Keep it private: it controls access to your bot.
The VPS runs OpenClaw and its tools. When you use a hosted model API, the model provider performs inference, so a GPU on the VPS is not required for that arrangement. Browser tasks, downloads and concurrent work still consume server resources. Choose capacity for your workload and allow room for the Docker image, logs and updates; a single RAM figure is not a guarantee for every task.
2. Connect your Hetzner project
In Hetzner Console, open the project that should own the server. Go to Security, then API tokens, and generate a token with Read & Write permissions. Copy it when Hetzner displays it; you cannot view the secret again after closing that window. This token allows resource changes in that project, so store it securely and enter it only in the intended setup form.
In ClawKickstart, open Create Instance and choose Hetzner Cloud. Paste the project token and select Verify token. Once verification succeeds, choose an instance name, location and server type from the available options. Check the current price and capacity before creating the server. Hetzner bills the infrastructure to your Hetzner account.
3. Select OpenClaw and provision the server
Choose OpenClaw explicitly in the framework selector. Hermes is the recommended default for most new deployments, but this guide is for OpenClaw. Review the server details, then create the instance and follow its provisioning progress in the dashboard. ClawKickstart prepares the host and installs the OpenClaw runtime through its Docker deployment path.
Wait for the instance to reach its running state before completing Agent Setup. Provisioning prepares the server; it does not by itself prove that the model key or Telegram bot can answer a message. If provisioning fails, open Recent activity and review the operation output before trying another deployment.
4. Configure the model and Telegram
Open Agent Setup for the running OpenClaw instance. Choose your provider, enter the model reference and supply the required API credentials. Use the provider’s exact model identifier rather than a marketing name. If you choose an authentication option that requires signing in on the host, complete that flow instead of assuming a normal API key is interchangeable.
For the Telegram path, open the verified BotFather account in Telegram, send /newbot and follow its prompts. Copy the bot token into the Telegram field in Agent Setup. The token is different from your Telegram username or personal numeric user ID. The personal user ID field belongs to the Hermes setup path; OpenClaw does not require it in this form.
Choose any optional personality or starter pack you want, review the settings and hatch the agent. Follow the hatch operation until it completes. OpenClaw also has a WhatsApp linking flow in the dashboard, but it requires separate channel configuration and QR pairing; use the dedicated WhatsApp guide rather than entering WhatsApp details into the Telegram fields.
5. Verify a real conversation
- Open your new bot in Telegram, press Start if prompted and send a short message such as “Reply with a sentence confirming you can hear me.”
- Confirm that the bot returns a model-generated reply after hatch completes.
- Try one small task representative of how you plan to use the agent before adding more integrations.
- Keep access to your Hetzner and model-provider accounts so you can monitor usage, manage credentials and recover the deployment.
If no reply arrives, run Diagnostics and review Recent activity. First separate server reachability and gateway health from provider authentication and messaging configuration. An invalid model name, exhausted provider credit or an incorrect bot token can prevent replies even when the server is running. Correct the relevant field and use the dashboard’s reconfiguration flow before testing again.
Keep gateway access private
You do not need to expose the gateway to the internet to chat through Telegram. ClawKickstart’s OpenClaw deployment restricts the published gateway ports to localhost. If you need direct gateway administration, use a private access method such as an SSH tunnel and follow the official remote-access documentation. A public domain, reverse proxy and TLS certificate are separate configuration decisions.
Your agent configuration and local state live on your VPS, but hosted model APIs and messaging services still process the data you send through them. Back up configuration and important agent state before major updates or changes. Optional Care adds monitoring features after deployment; it does not replace backups or the first successful conversation check.
Already have a VPS?
Choose the existing-server path in Create Instance instead of creating another Hetzner server. Have its public IP address, SSH username and port ready, then follow the dashboard’s SSH-key instructions. Back up an existing server before installation: the bootstrap changes host packages, firewall rules and the agent runtime. The current OpenClaw bootstrap is Ubuntu-focused, so check compatibility rather than assuming any Linux distribution will work.
