2026 OpenClaw cloud deployment: one-click image vs source build
You can either use a cloud provider’s OpenClaw/Claw image for a quick setup (often 3–5 minutes to reach the console) or install Node 22 and openclaw via npm/pnpm yourself (about 20–30 minutes, full control). One-click images are best for quick validation; source install suits teams that need a fixed stack. In 2026 most new users start with the image, then consider moving to self-hosted or a hosted remote Mac if they need long-term stability.
Requirements: Node 22, port 18789, firewall and security group
All of the following must be in place or the console will not be reachable.
| Item | Requirement | Notes |
|---|---|---|
| Node.js | ≥ 22.0.0 | Older versions can cause install or runtime errors; use nvm install 22 && nvm use 22 |
| Port | TCP 18789 | Default OpenClaw console/gateway port; must allow inbound in instance security group/firewall |
| OS and resources | Linux Ubuntu 20.04+ / macOS / Windows WSL2; min 2 cores, 2GB RAM, 20GB disk | Images usually meet this; self-hosted must verify |
On Tencent Cloud, Aliyun, etc., add an inbound rule: protocol TCP, port 18789, source 0.0.0.0/0 (or restrict by IP). Otherwise http://server-ip:18789 will time out.
Install issues: Node version, npm hang/timeout, mirror and pnpm
Common fixes for 2026:
# 1. Node version (must be >= 22)
node -v
nvm install 22
nvm use 22
# 2. npm mirror (avoid hang/timeout)
npm config set registry https://registry.npmmirror.com
# 3. Optional: use pnpm
npm install -g pnpm
pnpm add -g openclaw@latest
# 4. Official one-line install (all platforms)
curl -fsSL https://openclaw.ai/install.sh | bash
Node too old: Run node -v; if < 22, install Node 22 via nvm or official package.
npm install hangs or times out: Set a mirror (e.g. npmmirror.com) or switch to pnpm.
Docker container won’t start: Check if 18789 is in use (netstat -an | grep 18789) and whether the image was pulled completely.
Console not loading, Telegram/Feishu not replying
Console (http://IP:18789) times out or refused: In almost all cases the firewall is blocking 18789. In the cloud console, open the instance’s security group and add an inbound rule for TCP 18789; on your own Linux, check firewalld, ufw, or iptables. Restart OpenClaw after changing rules.
Telegram or Feishu bot not replying: Confirm OpenClaw is running (openclaw status or process list). Check the channel Token and bot settings. Use openclaw logs --tail 50 for connection or auth errors. You can enable multiple channels; history is per channel.
Multi-platform and offline (Ollama)
OpenClaw can expose the same AI to several channels (Telegram, Discord, Feishu, etc.) at once; history is isolated per channel. Skills and some channels need network; with Ollama you can run core chat offline. For models: daily chat (Kimi, MiniMax), coding (Claude Sonnet, GPT-4), cost-sensitive (DeepSeek, Qwen), privacy (Ollama local).
FAQ and when to use cloud image vs remote Mac
- Quick validation or short trial: Use the provider’s OpenClaw image; open 18789 and set API keys.
- 7×24 stable runtime without managing cloud firewall and OS: Consider moving OpenClaw to a hosted remote Mac (e.g. SFTPMAC) for availability, clear directories and permissions, and multi-platform access.
- Already on Apple/Mac workflows: A remote Mac aligns better with compatibility and file/desktop collaboration and simplifies backup and upgrade.
Cloud images are good for getting started; when usage shifts to always-on, multi-channel, and clear permissions, a hosted remote Mac often reduces ops and improves stability.
OpenClaw console not accessible?
Most often the firewall or security group does not allow port 18789. In the cloud console add an inbound rule for TCP 18789; on Linux check firewalld/ufw or iptables.
npm install hangs or times out?
Set a mirror, e.g. npm config set registry https://registry.npmmirror.com, or use pnpm. Ensure Node >= 22.0.0.
When to use cloud image vs remote Mac for OpenClaw?
Cloud image suits quick validation or short-term trial; for 7×24 stable runtime, Apple ecosystem, and clear directory/permission, use a hosted remote Mac (e.g. SFTPMAC).
Self-hosting OpenClaw on a cloud VM gets you running quickly, but long-term you deal with security groups, OS updates, and backups. If you prefer to focus on automation rather than infra, consider running OpenClaw on SFTPMAC’s remote Mac nodes: we handle port 18789, directory and permissions, and multi-platform access so you can focus on models and channels.