OpenClaw Installation Guide — All Platforms (Linux, macOS, Unraid)

⚠️ The Problem

Users encounter various installation issues when setting up OpenClaw, including: 1. Gateway won't start or is unreachable: ``


Gateway: unreachable (connect failed: connect ECONNREFUSED 127.0.0.1:18789)



2. **Ollama "address already in use" error:**


Error: listen tcp 127.0.0.1:11434: bind: address already in use



3. **Model keeps forgetting context / "Hello, who am I?" every few messages**

4. **NPM ENOENT errors during installation:**


npm ERR! enoent ENOENT: no such file or directory



5. **Small models producing garbage output** (raw TTS instructions, broken formatting)

6. **Dashboard showing "unauthorized: gateway token missing":**


disconnected (1008): unauthorized: gateway token missing (open a tokenized dashboard URL or paste token in Control UI settings)



7. **systemd service name confusion:**


Failed to enable unit: Unit file clawdbot.service does not exist.

🔍 Why This Happens

These issues stem from different root causes: 1. ECONNREFUSED — The gateway daemon isn't running. OpenClaw requires the gateway to be started before you can interact with it. 2. Ollama "address in use" — This is actually normal! It means Ollama is already running as a background service. You don't need to start it again. 3. Context amnesia — The model's context window is too small, OR the model itself is too small (0.5b-3b models can't maintain coherent conversations). The Ollama docs recommend 64k context, but 128k may be the minimum that works reliably. 4. NPM ENOENT — Usually means you're in the wrong directory, node_modules is corrupted, or the file has a hidden extension like package.json.txt. 5. Garbage output from small models — Models under 7B parameters struggle with system prompts and complex instructions. The 0.5b and 3b models are essentially unusable for agent work. 6. Gateway token missing — The dashboard requires authentication. You need to open the tokenized URL from the CLI or configure the token in settings. 7. systemd naming — The service is named clawdbot-gateway.service, not clawdbot.service.

✅ The Fix

## Fresh Installation

First, ensure you have the prerequisites installed:

bash

# Check Node.js version (need 22+)node --version# Install pnpm if you don't have itnpm install -g pnpm

Clone and build OpenClaw:

bash

git clone https://github.com/clawdbot/clawdbot.gitcd clawdbotpnpm installpnpm build

Run the onboarding wizard:

bash

pnpm clawdbot onboard

This walks you through model setup, workspace configuration, and channel connections (WhatsApp, Telegram, Discord, etc.).

## Starting the Gateway

The gateway must be running for OpenClaw to work:

bash

# Start the gatewayopenclaw gateway start# Check statusopenclaw gateway status# Enable auto-start on boot (correct service name!)sudo systemctl enable clawdbot-gateway

Note: The service is clawdbot-gateway.service, NOT clawdbot.service.

## Ollama Setup

If you see "address already in use" when running ollama serve, that's fine — Ollama is already running. Verify it's working:

bash

# Check running modelsollama ps# List installed modelsollama list# Test the APIcurl http://127.0.0.1:11434/api/tagscurl http://127.0.0.1:11434/v1/models

Important: Models don't load into RAM until something calls them. Test with:

bash

openclaw chat --model ollama/deepseek-r1:14b "hello"

Make sure you've pulled the model first:

bash

ollama pull deepseek-r1:14b

## Fixing Context Amnesia

If your bot keeps forgetting who it is every 2-3 messages, the context window is too small or the model is too weak.

Minimum viable models for CPU: - Qwen2.5 7B - Llama3 8B

For reliable performance: 14B+ models recommended

Check your Ollama model's context settings. The default 64k may not be enough — try 128k if you have the VRAM:

bash

# Check VRAM usage while model is loadednvtop

## Fixing NPM ENOENT Errors

bash

# Make sure you're in the project rootcd /path/to/clawdbotls package.json  # Should show the file# Clear and reinstallrm -rf node_modules package-lock.jsonpnpm install

On Windows, check if the file has a hidden extension:

powershell

ls -la | grep package# If it shows package.json.txt:ren "package.json.txt" package.json

## Complete Uninstall & Fresh Start

bash

# Stop and remove the daemonopenclaw daemon stopopenclaw daemon uninstall# Delete all OpenClaw datarm -rf ~/.clawdbot# Optional: delete workspacerm -rf ~/clawd# Fresh installgit clone https://github.com/clawdbot/clawdbot.gitcd clawdbotpnpm installpnpm buildpnpm clawdbot onboard

## Debugging Tips

Run these commands to diagnose issues:

bash

# Deep status checkopenclaw status --deep# Watch logs in real-timeopenclaw logs --follow# Check versionopenclaw --version

🔥 Your AI should run your business, not just answer questions.

We'll show you how.$97/mo (going to $197 soon)

Join Vibe Combinator →

📋 Quick Commands

Command	Description
openclaw gateway start	Start the gateway daemon
openclaw gateway status	Check if gateway is running
sudo systemctl enable clawdbot-gateway	Enable auto-start on boot
ollama ps	List currently running Ollama models
ollama list	List all installed Ollama models
ollama pull <model>	Download a model
openclaw status --deep	Deep diagnostic check
openclaw logs --follow	Watch logs in real-time
curl http://127.0.0.1:11434/api/tags	Test Ollama API
rm -rf node_modules && pnpm install	Clean reinstall dependencies

OpenClaw Installation Guide — All Platforms (Linux, macOS, Unraid)

⚠️ The Problem

🔍 Why This Happens

✅ The Fix

📋 Quick Commands

Related Issues

📚 You Might Also Like

Set Up OpenClaw for Beginners

30-Minute Setup Guide

Zoom

Self-Hosted AI Assistant: The Complete 2026 Guide

Still stuck?