Why self-host OpenClaw on a VPS
Most consumer AI assistants are cloud services: your messages, your contacts and your keys pass through a third party's infrastructure, in exchange for a monthly subscription in dollars. OpenClaw reverses this model. It is a gateway that you run yourself: it links your messaging apps to an LLM provider you choose, and everything passes through your VPS. You decide which channels to enable (WhatsApp, Telegram, Discord, Signal, Matrix…), which model to use (OpenAI, Anthropic, or a local model via Ollama) and which skills the assistant may execute. On a VPS, you benefit from a stable IP, permanent availability and a fixed cost — the assistant stays reachable 24/7 without leaving a personal machine switched on.
The concrete benefits of a self-hosted OpenClaw
- Data sovereignty: the gateway and your keys stay on your VPS, nothing is entrusted to a cloud assistant.
- Controlled cost: no subscription in USD; with a local model via Ollama, zero API cost.
- Multi-messaging: a single assistant reachable from WhatsApp, Telegram, Discord, Slack and 50+ channels.
- Available 24/7: a VPS stays on permanently, unlike a personal workstation.
- Extensible: add skills and MCP tools according to your needs.
Onboarding: LLM provider and channels
OpenClaw is not "ready to use" without a configuration step, and that is deliberate: on first launch, an onboarding wizard asks for your LLM provider and generates the gateway token written to the .env file. At this point you choose whether to use a remote API (OpenAI, Anthropic) or a local model served by Ollama on the same server. Then comes the linking of channels: each messaging app (WhatsApp, Telegram, Discord…) has its own pairing wizard from the Control UI. It is this step that turns the gateway into a genuinely reachable assistant.
Deploy OpenClaw with Docker
Prepare the VPS
On an Ubuntu 22.04 VPS (2 GB RAM minimum, 4 GB recommended), install Docker and Docker Compose. Plan for 8 to 16 GB if you intend to run a local LLM model via Ollama on the same machine.
Launch the gateway
Use the official image ghcr.io/openclaw/openclaw:latest rather than building it locally. Deployed via Docker Compose, the gateway listens on port 18789 (Control UI + gateway).
Complete the onboarding
On first start, follow the onboarding wizard: enter your LLM provider and your key; the gateway token is generated and stored in .env.
Connect your messaging apps
From the Control UI, link the desired channels by following the pairing wizard for each integration (WhatsApp QR code, Telegram bot token, etc.).
Secure access with HTTPS
Place an Nginx or Caddy reverse proxy with a Let's Encrypt certificate in front of port 18789. Never expose the Control UI in the clear on the Internet.
To preserve your configuration and the channel pairing history, back up OpenClaw's Docker volumes (the config and workspace directories) as well as the .env file that contains the gateway token. A redeployment then restarts without having to re-pair all your messaging apps.
The official documentation
For advanced configuration and tool-specific options, refer to the official OpenClaw documentation. This guide covers bringing it online on a VPS; the vendor's docs remain the reference for fine-tuning, major updates and specific use cases.
