OpenClaw Gateway Crashes or Won't Start — Complete Fix Guide

## Quick Diagnosis

Run these commands first to understand the problem:

# Check overall statusopenclaw status --all# Check gateway specificallyopenclaw gateway status# Check if models are configuredopenclaw models status# Check channel connectionsopenclaw channels status --probe

What to look for:

- Gateway: unreachable → Gateway not running

- Gateway service: Scheduled Task not installed → Windows service issue

- Models: no default model → API key issue

- Channels: disconnected → Channel-specific issue

## Fix Gateway Not Running (1006 Errors)

The 1006 error means the WebSocket closed unexpectedly — usually because the gateway isn't running.

Step 1: Check if gateway is running:

openclaw gateway status

Step 2: Start the gateway:

openclaw gateway start

Step 3: If it was running but crashed, check logs:

openclaw gateway logs --tail 50

Step 4: Look for port conflicts:

# macOS/Linuxlsof -i :18789# Windows PowerShellGet-NetTCPConnection -LocalPort 18789

## Fix Windows Gateway Issues

On Windows, the gateway runs as a Scheduled Task that needs to be installed:

Step 1: Install the Windows service:

openclaw gateway install

Step 2: Check service status:

Get-ScheduledTask -TaskName "OpenClaw Gateway"

Step 3: Start the service:

Start-ScheduledTask -TaskName "OpenClaw Gateway"

Step 4: Verify it's running:

openclaw gateway status

Common Windows issue — CLI syntax:

# This is WRONG on some versions:openclaw gateway restart# error: too many arguments for 'gateway'# Try this instead:openclaw gateway call gateway.restart# Or stop and start:openclaw gateway stopopenclaw gateway start

## Fix 1008 Token Mismatch Errors

The remote token must exactly match the auth token:

Step 1: Check current token configuration:

openclaw config get gateway.auth.tokenopenclaw config get gateway.remote.token

Step 2: If they don't match, set them to be identical:

openclaw config set gateway.remote.token "your-exact-token-here"

Step 3: Apply config and restart:

openclaw config applyopenclaw gateway stopopenclaw gateway start

Step 4: Also check browser localStorage:

Open DevTools (F12) → Application → Local Storage → look for openclaw.control.settings.v1. The token there must match your config.

## Fix "No Response" Issues (Gateway Running but Bot Silent)

If the bot sends scheduled messages but doesn't respond to DMs, the gateway is running but not processing inbound messages.

Step 1: Watch logs in real-time and send a test message:

openclaw logs --follow# Now send "ping" to your bot in Telegram/Discord

What to look for:

- Nothing logged when you DM → Receiver is dead/misconfigured (webhook or polling issue)

- Message logged but "blocked/unauthorized/pairing" → Access control issue

- Message logged but model errors → Provider/API key issue

Step 2: Fix Telegram webhook conflicts:

# Check if a webhook is setcurl https://api.telegram.org/bot<YOUR_BOT_TOKEN>/getWebhookInfo# If webhook_url is set, delete it for long-polling:curl https://api.telegram.org/bot<YOUR_BOT_TOKEN>/deleteWebhook

Step 3: Check access controls:

# Check if you need to approve pairingopenclaw devices list# Approve if pendingopenclaw devices approve <CODE># Check allowlist (for Telegram)openclaw config get channels.telegram.allowFrom# This should contain your numeric Telegram user ID

To get your Telegram user ID:

Message @userinfobot on Telegram — it will tell you your numeric ID.

## Fix Context Overflow / Memory Issues

Long-running sessions can hit context limits:

Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

Solution 1: Reset the session:

# In chat, send:/reset

Solution 2: Enable auto-compaction:

Add to your config:

{  "agent": {    "compaction": {      "enabled": true,      "threshold": 50000    }  }}

Solution 3: Use a larger context model:

Claude 3.5 Sonnet has 200k context, GPT-4-turbo has 128k. Switch if you're hitting limits.

## Fix Model/Provider Errors

If the gateway runs but the model fails:

Step 1: Check model status:

openclaw models status

Step 2: Verify API key is set:

# For Anthropicopenclaw config get models.anthropic.apiKey# For OpenAIopenclaw config get models.openai.apiKey

Step 3: Test the model directly:

openclaw models test anthropic/claude-sonnet-4-20250514

## Fix DeepSeek and Other Custom Models

DeepSeek isn't in the default model list. Configure it manually:

Step 1: Configure DeepSeek in openclaw.json:

{  "models": {    "openai": {      "apiKey": "your-deepseek-api-key",      "baseUrl": "https://api.deepseek.com"    }  },  "agent": {    "defaultModel": "openai/deepseek-chat"  }}

Step 2: Apply and restart:

openclaw config applyopenclaw gateway restart

## Full Gateway Log Analysis

When all else fails, get the full logs:

# Last 100 linesopenclaw gateway logs --tail 100# Follow in real-timeopenclaw logs --follow# With verbose outputDEBUG=* openclaw gateway start

## Emergency Reset

If the gateway is completely broken:

# Stop everythingopenclaw gateway stop# Clear gateway state (keeps config)rm -rf ~/.openclaw/gateway-state/# Fresh startopenclaw gateway start

On Windows:

openclaw gateway stopRemove-Item -Recurse -Force $env:USERPROFILE\.openclaw\gateway-state\openclaw gateway start