How to deploy OpenClaw

OpenClaw crossed 376,000+ GitHub stars by June 2026, and the most common question new users ask is simply: how do I deploy it? There are two honest answers. You can self-host it with Docker on a VPS and own every moving part, or you can let a managed platform run it for you in an isolated container. This guide walks both paths end to end so you can pick the one that fits your time and your ops appetite.

OpenClaw is a Node.js application distributed as a Docker image from github.com/openclaw/openclaw. Whichever path you take, the core ingredients are the same — what changes is how much of the plumbing you own.

What you need

Before you deploy OpenClaw either way, line up these four things:

• A server — a VPS for the DIY route (Hetzner, Contabo, DigitalOcean, etc.), or none at all if you go managed.
• Docker — the container runtime that runs the OpenClaw image. On the managed path this is already there.
• A model API key — OpenClaw is BYOK, so you bring a key from a provider such as OpenAI, Anthropic, Google, Groq, or any of the 13 providers VibeOpenClaw supports.
• A channel bot token — a Telegram, Discord, or Slack bot token so people can actually talk to the agent.

Path 1 — DIY with Docker

The self-hosted route gives you full control of the filesystem and the box. It also makes you the operator. Here is the realistic sequence, not the marketing version:

1. Rent a VPS. Spin up a small Linux instance — 2 GB of RAM is a sane minimum for a single OpenClaw agent. Note the public IP, then SSH in. You now own this machine, including its security updates.
2. Install Docker. Install the Docker engine and confirm the daemon is running. If you want multiple agents or supporting services, you may add Docker Compose here too.
3. Pull and run OpenClaw. Pull the OpenClaw image (it is a Node.js app under the hood) and start a container. Map a port and a volume so config and conversation state survive restarts.
4. Configure your channel token and model key. Pass your Telegram/Discord/Slack bot token and your model provider API key into the container as environment variables or a config file. Treat both as secrets — never bake them into an image or commit them.
5. Set up SSL and a reverse proxy. Put something like Nginx or Caddy in front, point a domain at the box, and get TLS certificates issued and auto-renewing. This is also where you handle port forwarding and firewall rules so the right ports are open and the rest are closed.
6. Keep it running 24/7. Add a restart policy (for example --restart=always) so the container recovers from crashes and reboots, set up monitoring so you find out when it falls over, and schedule backups of your volumes. From here on, certificate renewals, OS patches, and restarts are your job.

None of these steps is exotic, but together they are real, ongoing work. If you enjoy owning the box and have done this before, the DIY path is the cheapest in dollars — a VPS runs roughly $5–$15/month. If SSH, reverse proxies, and 2 a.m. restart scripts sound like a chore, keep reading.

Path 2 — One-click on VibeOpenClaw

The managed path skips every server step above. VibeOpenClaw runs OpenClaw for you in a Docker-isolated container with SSL, restarts, and updates already handled. Here is the whole flow:

1. Sign up for a plan — Pro ($24/mo) runs one OpenClaw agent with 2 GB of RAM on Telegram and Discord; Premium ($48/mo) runs up to three OpenClaw or Hermes agents with 4 GB each and adds Slack.
2. Add your provider key on the API Keys page. Paste a key from any of the 13 supported providers — OpenAI, Anthropic, Google, Groq, xAI, Mistral, DeepSeek, Together, Fireworks, Perplexity, OpenRouter, Cohere, or NVIDIA. Keys are encrypted at rest with AES-256-GCM, and we never mark up inference.
3. Click New Agent → OpenClaw to start a fresh deploy.
4. Pick your model from the provider key you just added.
5. Paste your Telegram or Discord bot token (Slack too, on Premium) so the agent has a channel to talk through.
6. Click Create. Your agent is live in about 30 seconds, running in its own Docker-isolated container with SSL and automatic restarts handled for you.

That is the entire deploy. No VPS to rent, no Docker to install, no reverse proxy to configure, and no restart policy to write. When you want a second agent or want to add Hermes alongside OpenClaw, you repeat the New Agent step.

Which path?

The verdict comes down to what you value. Choose the DIY Docker path if you want full filesystem control, already run your own infrastructure, and treat the ops work as a feature rather than a tax. Choose the managed path if you want OpenClaw live in 30 seconds, isolated per agent, with encrypted keys, SSL, and restarts handled — and you would rather spend your time using the agent than babysitting it. For most people, managed wins; for tinkerers and infra teams, DIY is a fair trade.

DIY vs managed at a glance

Want to go deeper? See our managed OpenClaw hosting page, compare the agents in OpenClaw vs Hermes, or weigh providers in the best OpenClaw hosting providers in 2026.