OpenClaw Installation and MYAI168 Setup Guide — macOS

Goal of this guide: install OpenClaw (an open-source AI agent that runs on your own machine, nicknamed the "lobster" 🦞) on macOS and connect it to your own MYAI168 API, using Claude Opus 4.8 (with gpt-5.6-sol as an optional backup).

Endpoints and Model

Anthropic native endpoint    https://www.myai168.com/en/api/anthropic
OpenAI-compatible endpoint   https://www.myai168.com/en/api/openai/v1
Main model                   claude-opus-4-8 (via the Anthropic endpoint)

Important Security Notes

• The lobster has very broad system permissions (reading files, running commands, sending messages, browsing the web). Do not install it on a primary machine holding lots of sensitive data; if possible use a dedicated machine or a clean account.
• Do not expose the admin console (Control UI) to the public internet; keep it on localhost or behind a Tailscale / SSH tunnel.
• Always install the latest version (older versions had serious security vulnerabilities).
• Early token usage can be high; set a usage cap in the MYAI168 dashboard.

Prerequisites

Node.js 22.22.3+, 24.15+, or 25.9+ (Node 24 is the official default target and the recommended choice).

Check Node (if missing, brew install node):

node --version

Install the lobster (official script, or via npm if you already have Node):

curl -fsSL https://openclaw.ai/install.sh | bash
# or: npm i -g openclaw@latest

If you hit an npm permission error (EACCES), point the global directory to your home directory:

mkdir -p ~/.npm-global
npm config set prefix ~/.npm-global
echo 'export PATH=~/.npm-global/bin:$PATH' >> ~/.zshrc
source ~/.zshrc
npm install -g openclaw@latest

Verify:

openclaw --version

openclaw onboard

Answer in order:

• Security disclaimer → Yes
• Setup mode → QuickStart
• Model/auth provider → Skip for now (we will connect a custom API by editing the config file later)
• Messaging platforms (Telegram / WhatsApp…) → skip them all
• Generate gateway token? → Yes
• Tighten permissions? → Yes
• Create session store dir? → Yes
• Disable unavailable skills? → No (keeping them does no harm; they start working once their dependencies are installed)

Set the Gateway to local mode:

openclaw config set gateway.mode local

mkdir -p ~/.openclaw
nano ~/.openclaw/openclaw.json

Paste in (replace "your-key" with your real key; use nano so TextEdit does not turn straight quotes into curly quotes and break the JSON):

{
  models: {
    mode: "merge",
    providers: {
      myai168: {
        baseUrl: "https://www.myai168.com/en/api/openai/v1",
        apiKey: "your-key",
        api: "openai-completions",
        models: [
          { id: "gpt-5.6-sol", name: "GPT-5.6 Sol", input: ["text"],
            contextWindow: 200000, maxTokens: 32000 }
        ]
      },
      myai168_claude: {
        baseUrl: "https://www.myai168.com/en/api/anthropic",
        apiKey: "your-key",
        api: "anthropic-messages",
        models: [
          { id: "claude-opus-4-8", name: "Claude Opus 4.8", input: ["text"],
            contextWindow: 200000, maxTokens: 32000 }
        ]
      }
    }
  },
  agents: {
    defaults: {
      model: { primary: "myai168_claude/claude-opus-4-8" },
      models: {
        "myai168_claude/claude-opus-4-8": {},
        "myai168/gpt-5.6-sol": {}
      }
    }
  }
}

(In nano: save with Ctrl+O, Enter; exit with Ctrl+X.)

Key Points

• baseUrl has a lowercase u; the api field takes openai-completions / anthropic-messages.
• A provider must be defined AND listed in the agents.defaults.models allowlist, otherwise you get model not allowed.
• primary is the default main model; to default to gpt-5.6-sol, change it to "myai168/gpt-5.6-sol".
• Set the anthropic endpoint baseUrl to https://www.myai168.com/en/api/anthropic (without /v1) — OpenClaw's Anthropic client appends /v1 to the base URL automatically (per the official docs), so adding /v1 yourself produces a duplicated /v1/v1 path that fails.

Confirm the models were loaded:

openclaw models list

You should see your models, with yes in the Auth column and the main model tagged default / configured. (If a claude-cli/... entry also appears, it was auto-detected from a local Claude CLI install — ignore it.)

Launch test:

openclaw

Once in the chat UI (if the built-in OpenClaw setup agent — formerly called Crestodian — appears first, it hands you over to the main agent; the status bar at the bottom should show myai168-claude/claude-opus-4-8), ask a math question:

What is 123456 times 789? Reply with the number only.

If it answers 97406784, the connection works. (Do not verify with "who are you" — model self-identification is unreliable.)

You can also try chatting in the dashboard:

openclaw dashboard      # opens http://127.0.0.1:18789/ in the browser

• command not found: openclaw → PATH not in effect. Run source ~/.zshrc or reopen the terminal. (On Linux: source ~/.bashrc.)
• npm ... EACCES → global directory permission problem; change the prefix to ~/.npm-global (see "1. Prerequisites and Installation").
• doctor reports a JSON syntax error (after manual editing) → the editor turned straight quotes into curly quotes. Edit with nano, or disable smart quotes and paste again.
• model not allowed → the model is not on the allowlist; add it under agents.defaults.models. (On Linux, cc-switch usually takes care of this.)
• Replies fail with model-related errors → the model id or endpoint does not match the provider's actual values; compare and fix.
• Switch the main model → change agents.defaults.model.primary in the config file. (On Linux: switch the enabled provider in cc-switch.)

Update:

openclaw update
# or: npm update -g openclaw

Security audit:

openclaw security audit --deep

Rotate the API Key

Update the apiKey in ~/.openclaw/openclaw.json and restart OpenClaw. (On Linux: update the API Key on the cc-switch card and Enable it again.)

Config File Location

~/.openclaw/openclaw.json (the same directory also holds state/, logs/, etc.)

Official docs: https://docs.openclaw.ai