OpenClaw quickstart: install, connect Codex, and get your first chat reply

Install OpenClaw, finish Codex OAuth, open the dashboard, and send your first successful chat. This beginner quickstart skips Telegram channels, skills, and API keys so you can get a real first win fast.

1) Install OpenClaw (copy/paste)

curl -fsSL https://openclaw.bot/install.sh | bash

iwr -useb https://openclaw.ai/install.ps1 | iex

1.1 What you should see (real example output)

Your output will differ, but it should look like this:

🦞 OpenClaw Installer
Type the command with confidence—nature will provide the stack trace if needed.

✓ Detected: linux
✓ Node.js v25.5.0 found
✓ Git already installed
→ Configuring npm for user-local installs...
✓ npm configured for user installs
→ Installing OpenClaw 2026.1.29...
✓ OpenClaw installed successfully (2026.1.29)!
Starting setup...

1.2 Fix the common PATH warning (if you see it)

If the installer prints a warning like:

PATH warning: missing npm global bin dir: /home/it/.npm-global/bin
This can make openclaw show as "command not found" in new terminals.

Do exactly what it says: add the printed export PATH=... line into your shell config, then open a new terminal:

• bash: ~/.bashrc
• zsh: ~/.zshrc

Official reference: Node.js / npm PATH sanity

1.3 Verify

openclaw --help

2) Run onboarding (do exactly these choices)

openclaw onboard --install-daemon

2.1 Security warning

Choose Yes:

I understand this is powerful and inherently risky. Continue?
Yes / No

2.2 Onboarding mode

Choose QuickStart:

Onboarding mode
QuickStart (Configure details later via openclaw configure.)
Manual

2.3 Model/auth provider

Choose OpenAI → OpenAI Codex (ChatGPT OAuth):

Model/auth provider
OpenAI (Codex OAuth + API key)
...

OpenAI auth method
OpenAI Codex (ChatGPT OAuth)
OpenAI API key

2.4 Codex OAuth on a VPS (copy/paste flow)

If you are on a remote/VPS, OpenClaw will show you an OAuth URL to open in your local browser. After signing in, you must paste the redirect URL back into the terminal.

Open this URL in your LOCAL browser:
https://auth.openai.com/oauth/authorize?response_type=code&client_id=...&redirect_uri=http%3A%2F%2Flocalhost%3A1455%2Fauth%2Fcallback&...

http://localhost:1455/auth/callback?code=...&scope=...&state=...

Paste the redirect URL (or authorization code)

After success you should see:

OpenAI OAuth complete

2.5 Default model

Choose Keep current (recommended). Example:

Default model set to openai-codex/gpt-5.2
Default model
Keep current (openai-codex/gpt-5.2)

2.6 Channels

At “Select channel”, scroll to the bottom and choose Skip for now (this is the shortest path). You can add Telegram later.

Select channel (QuickStart)
...
Skip for now (You can add channels later via `openclaw channels add`)

Important: you may need to press Space to select “Skip for now”, then press Enter.

2.7 Skills / dependencies / API keys / hooks (skip everything for the fastest win)

Onboarding may ask about skills, installing dependencies, setting API keys, and enabling hooks. For the shortest path to first chat:

• Install missing skill dependencies: choose Skip for now.
• API keys for optional skills (Google Places / Gemini / Notion / ElevenLabs / etc): choose No.
• Hooks: choose Skip for now.

2.8 Linux (systemd): allow “lingering” if prompted

On Linux, QuickStart uses a systemd user service. If OpenClaw offers to enable lingering, allow it (may require sudo). This prevents the gateway from being killed on logout/idle.

3) Onboarding completion: copy the tokenized dashboard link

At the end, OpenClaw prints a “Dashboard ready” section. This is the moment you’ve been waiting for.

Dashboard ready

Dashboard link (with token):
http://127.0.0.1:18789/?token=<YOUR_TOKEN>

No GUI detected. Open from your computer:
ssh -N -L 18789:127.0.0.1:18789 user@YOUR_SERVER_IP

Then open:
http://localhost:18789/
http://localhost:18789/?token=<YOUR_TOKEN>

3.1 If you see these warnings (they do NOT block the dashboard win)

• gateway closed (1008): pairing required — pairing is a safety feature. You can still proceed via the tokenized dashboard link above.
• Missing Control UI assets. Build them with pnpm ui:build — this matters if you’re building from source; if you installed via installer, you can usually ignore it and just open the dashboard link. If your UI is actually missing/broken, then follow the message.

Official references: Control UI · Remote access · Pairing

4) Open the dashboard + do the “hello” test

http://127.0.0.1:18789/?token=<YOUR_TOKEN>

ssh -N -L 18789:127.0.0.1:18789 user@YOUR_SERVER_IP

http://localhost:18789/?token=<YOUR_TOKEN>

5) Next: let OpenClaw configure the rest (copy/paste this prompt)

Now that you have a working chat + working auth, paste this into the dashboard and let OpenClaw generate a beginner-safe plan for Telegram/skills/etc:

You are OpenClaw. I already completed install + onboarding (Codex OAuth) and I can chat in the dashboard.

My goal: configure the remaining setup safely, as a beginner.

Ask me 5-8 clarifying questions first (OS, local vs VPS, whether I want Telegram, whether I want Docker, and my security posture).
Then produce a step-by-step checklist.
Rules:
- One step at a time.
- After each step, ask me to paste the output (or say DONE).
- Never ask me to paste tokens into a public/group chat.
- If you are unsure, say UNKNOWN and ask for the missing info.
Finally, link me to the exact OpenClaw CLI commands I should run for each step.

Bookmark these (save for later)