Setting Up OpenClaw on exe.dev with Discord

By Brian Douglas on Feb 2, 2026. Originally published on DEV.to.
Setting Up OpenClaw on exe.dev with Discord

I'll be honest—when I first heard about Clawdbot (now rebranded as OpenClaw) a few weeks back, I had that immediate security alarm going off in my head. An AI agent running with Discord permissions, making API calls, potentially accessing who knows what? As someone who's spent years in developer experience, the threat model here was... concerning.

But then I started seeing folks in the community actually using it—spinning up Digital Ocean droplets, deploying to Railway.

So I decided to try it, but on my terms: isolated, observable, and disposable. Which is why I chose exe.dev, but this can work on the platform of your choice.

Why exe.dev?

I've known about exe.dev for months—it's this clever service that gives you ephemeral VMs that you can spin up and tear down instantly. Perfect for experiments you're not 100% sure about. If OpenClaw does something weird or I misconfigure something, I just nuke the VM and start fresh. No damage to my local machine, no persistent infrastructure to maintain.

Think of it as a playground where mistakes are cheap.

💡 Pro Tip: Add Telemetry from Day One

Before you start, consider adding tapes as a proxy layer. It records every API call OpenClaw makes, giving you visibility into prompts, token usage, and agent behavior. The official guide has details, or follow the optional setup in Step 7 below. Trust me—you'll want this data when debugging why your agent did something unexpected.

What You'll Need

Before we dive in, grab these:

The Fast Path: Using the Template

Here's where exe.dev really shines. They have an OpenClaw template that handles most of the setup:

  1. Navigate to exe.new/openclaw
  2. Drop in your API keys when prompted
  3. Let Shelley (exe.dev's setup assistant) do the heavy lifting

I'm going to walk you through the manual setup though, because understanding what's actually happening is half the value here.

The Manual Route (More Learning, Same Result)

Create Your VM 

ssh exe.dev new
# You'll get a VM name like: curious-tesla-8432
ssh curious-tesla-8432.exe.xyz
Enter fullscreen mode Exit fullscreen mode

You're now inside an Ubuntu VM that exists just for you, accessible from anywhere.

Install OpenClaw

This is surprisingly straightforward:

# Get the system ready
sudo apt-get update
sudo apt-get install -y git curl jq ca-certificates openssl nginx

# Install OpenClaw via their script
curl -fsSL https://openclaw.bot/install.sh | bash

# Run the onboarding (we're skipping interactive prompts)
openclaw onboard --non-interactive --accept-risk
Enter fullscreen mode Exit fullscreen mode

That --accept-risk flag made me pause. It's OpenClaw's way of acknowledging that yes, you're running an AI agent with real permissions. I appreciate the directness.

Configure Nginx as a Reverse Proxy

OpenClaw runs on port 18789 internally, but we want to access it via standard web ports. This is where nginx comes in.

Edit /etc/nginx/sites-enabled/default:

server {
    listen 80 default_server;
    listen [::]:80 default_server;
    listen 8000;
    listen [::]:8000;

    server_name _;

    location / {
        proxy_pass http://127.0.0.1:18789;
        proxy_http_version 1.1;

        # WebSocket support - crucial for real-time updates
        proxy_set_header Upgrade $http_upgrade;
        proxy_set_header Connection "upgrade";

        # Standard proxy headers
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
        proxy_set_header X-Forwarded-Proto $scheme;

        # Long timeouts for agent tasks that might run a while
        proxy_read_timeout 86400s;
        proxy_send_timeout 86400s;
    }
}
Enter fullscreen mode Exit fullscreen mode

Apply the changes:

sudo systemctl restart nginx
Enter fullscreen mode Exit fullscreen mode

Connect Your Discord Bot

Now for the fun part. Edit ~/.openclaw/openclaw.json and add your Discord credentials:

{
  "channels": {
    "discord": {
      "enabled": true,
      "token": "YOUR_DISCORD_BOT_TOKEN"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Restart the gateway to pick up the changes:

openclaw gateway restart
Enter fullscreen mode Exit fullscreen mode

The Pairing Dance (This Tripped Me Up)

When I first tried to access the dashboard, I got this cryptic message:

disconnected (1008): pairing required

Turns out OpenClaw uses a device pairing system to prevent unauthorized access. Smart, but not immediately obvious.

Here's how to fix it:

# See what's waiting for approval
openclaw devices list
Enter fullscreen mode Exit fullscreen mode

You'll see something like:

Pending:
  - id: abc123
    browser: Chrome
    requested: 2 minutes ago
Enter fullscreen mode Exit fullscreen mode

Approve it:

openclaw devices approve abc123
Enter fullscreen mode Exit fullscreen mode

Now refresh your browser. You should be connected.

Access Your Dashboard

Your OpenClaw dashboard lives at:

https://<vm-name>.exe.xyz/?token=<your-gateway-token>
Enter fullscreen mode Exit fullscreen mode

Find your token in ~/.openclaw/openclaw.json under gateway.auth.token. I bookmarked this URL because I reference it constantly.

Verify Everything Works 

openclaw health
Enter fullscreen mode Exit fullscreen mode

You should see confirmation that Discord is connected and your agents are running:

Discord: ok (@yourbotname)
Agents: main (default)
Enter fullscreen mode Exit fullscreen mode

This is your health check. Anytime something feels off, run this first.

Optional: Route Through Tapes for Observability

Here's where my Continue background kicks in. I want to see what API calls OpenClaw is making, so I route everything through tapes—Anthropic's proxy tool for inspecting and recording API interactions.

Start tapes:

tapes serve \
  --provider anthropic \
  --upstream "https://api.anthropic.com" \
  --proxy-listen "0.0.0.0:8080" \
  --sqlite "./tapes.db"
Enter fullscreen mode Exit fullscreen mode

Update ~/.openclaw/openclaw.json:

{
  "providers": {
    "anthropic": {
      "baseUrl": "http://localhost:8080"
    }
  }
}
Enter fullscreen mode Exit fullscreen mode

Restart: openclaw gateway restart

Now every Claude API call is logged to tapes.db. You can inspect prompts, responses, token usage—everything. This level of observability is crucial when you're evaluating AI tooling.

Things That Surprised Me

  1. The pairing system actually works well - Once I understood it, I appreciated having explicit device approval
  2. Config persistence is fragile - Some OpenClaw commands will overwrite your config. Always backup ~/.openclaw/openclaw.json before running setup commands
  3. It's genuinely useful - Having a Discord bot that can reason about context and execute tasks is different from slash commands or simple bots

Troubleshooting Guide

"pairing required" won't go away 

openclaw devices list
openclaw devices approve <id>
Enter fullscreen mode Exit fullscreen mode

Refresh your browser after approving.

Discord shows offline

Gateway won't start 

openclaw gateway status  # Check what's wrong
openclaw logs           # See error details
openclaw gateway restart # Try restarting
Enter fullscreen mode Exit fullscreen mode

Quick Reference Commands

Command What It Does
openclaw health Connection status check
openclaw devices list Show paired/pending devices
openclaw devices approve <id> Approve a new device
openclaw gateway restart Restart the gateway process
openclaw doctor Run full diagnostics
openclaw logs View gateway logs

Final Thoughts

Running OpenClaw on exe.dev turned out to be the perfect middle ground. I got to experiment with AI agents in Discord without compromising my local environment, and I could observe everything happening via tapes.

Is it production-ready for your use case? That depends on what you're building. But as a tool for understanding how autonomous AI agents operate, how they handle context, and what the developer experience looks like? It's been genuinely educational.

The security concerns I had initially? Still valid. But now they're informed concerns rather than fear-based ones. And that's the difference between avoiding technology and understanding its tradeoffs.

Guide based on OpenClaw 2026.1.29 running on exe.dev. If you try this and run into issues, reach out—I'm probably debugging the same thing.