OpenClaw + ClashX: Smart Proxy Routing for Your Local AI Agent

What is OpenClaw

OpenClaw is the fastest-growing open-source project on GitHub in 2026, crossing 210K+ stars in under six months. Built by Austrian developer Peter Steinberger (founder of PSPDFKit) as a personal side project in late 2025, it was acquired by OpenAI in February 2026 and is now transitioning to the independent OpenClaw Foundation.

Unlike ChatGPT in a browser tab, OpenClaw is a locally-running autonomous AI agent. It lives on your machine, connects to messaging apps you already use (WhatsApp, Telegram, Discord, Slack, Signal, iMessage), and actually executes tasks — managing email, scheduling meetings, manipulating files, running shell commands, automating browsers — 24/7, even when you're asleep.

OpenClaw supports any LLM backend: OpenAI GPT, Anthropic Claude, local models (Llama, Phi), DeepSeek, and more. Your data stays local by default — nothing leaves your machine unless you explicitly send an API request.

Why Use a Proxy with OpenClaw

Even with unrestricted internet access, there are solid reasons to route OpenClaw traffic through ClashX:

Privacy: Separate Agent Traffic from Personal Browsing

OpenClaw makes dozens of API calls per hour — to LLM providers, messaging platforms, email servers, and web services. Without traffic separation, your ISP sees all of this mixed with your personal browsing. ClashX lets you route agent traffic through a dedicated path, keeping your AI usage private.

Performance: Dedicated Routing for Different APIs

Different LLM providers have servers in different regions. Routing OpenAI calls through a US proxy and Anthropic calls through a different path can reduce latency. ClashX's rule-based routing makes this trivial.

Travel: Pre-configured Access for Restricted Regions

If you travel to China, Russia, Iran, or other countries with internet restrictions, your always-on OpenClaw agent needs proxy access ready to go. Configure it once, and your agent keeps working no matter where you are.

Split-Routing: Keep Local Traffic Direct

• Local models (Llama, Phi via Ollama) — route DIRECT, no proxy needed
• DeepSeek API — accessible globally, route DIRECT to reduce latency
• OpenAI / Anthropic APIs — route through proxy for privacy or geo-optimization
• Messaging platforms — route based on your needs

ClashX Routing Configuration

First, download and install ClashX. Then configure split-routing rules for OpenClaw traffic.

1. Create a Dedicated AI Agent Proxy Group

OpenClaw maintains long-running connections to APIs and messaging platforms. A dedicated proxy group with stability-optimized settings prevents disruptions:

proxy-groups:
  - name: "🤖 AI Agent"
    type: url-test
    url: http://www.gstatic.com/generate_204
    interval: 600
    tolerance: 200
    proxies:
      - 🇺🇸 US-West
      - 🇺🇸 US-East
      - 🇯🇵 Japan

  - name: "🤖 AI Manual"
    type: select
    proxies:
      - 🤖 AI Agent
      - 🇺🇸 US-West
      - 🇯🇵 Japan
      - DIRECT

2. Split-Routing Rules

rules:
  # === Route through proxy ===
  # OpenAI
  - DOMAIN-SUFFIX,openai.com,🤖 AI Agent
  - DOMAIN-SUFFIX,oaiusercontent.com,🤖 AI Agent
  - DOMAIN-SUFFIX,oaistatic.com,🤖 AI Agent
  - DOMAIN-SUFFIX,chatgpt.com,🤖 AI Agent

  # Anthropic (Claude)
  - DOMAIN-SUFFIX,anthropic.com,🤖 AI Agent
  - DOMAIN-SUFFIX,claude.ai,🤖 AI Agent

  # Google AI
  - DOMAIN-SUFFIX,generativelanguage.googleapis.com,🤖 AI Agent
  - DOMAIN-SUFFIX,aistudio.google.com,🤖 AI Agent

  # === Direct (no proxy needed) ===
  # DeepSeek — globally accessible, lower latency direct
  - DOMAIN-SUFFIX,deepseek.com,DIRECT

  # Local model APIs (Ollama, LM Studio)
  - IP-CIDR,127.0.0.0/8,DIRECT
  - IP-CIDR,192.168.0.0/16,DIRECT

  # === Messaging platforms (optional proxy) ===
  # WhatsApp
  - DOMAIN-SUFFIX,whatsapp.com,🤖 AI Agent
  - DOMAIN-SUFFIX,whatsapp.net,🤖 AI Agent

  # Telegram
  - DOMAIN-SUFFIX,telegram.org,🤖 AI Agent
  - DOMAIN-SUFFIX,t.me,🤖 AI Agent
  - IP-CIDR,91.108.0.0/16,🤖 AI Agent,no-resolve
  - IP-CIDR,149.154.160.0/20,🤖 AI Agent,no-resolve

  # Discord
  - DOMAIN-SUFFIX,discord.com,🤖 AI Agent
  - DOMAIN-SUFFIX,discord.gg,🤖 AI Agent
  - DOMAIN-SUFFIX,discordapp.com,🤖 AI Agent

  # Slack
  - DOMAIN-SUFFIX,slack.com,🤖 AI Agent
  - DOMAIN-SUFFIX,slack-edge.com,🤖 AI Agent

  # Signal
  - DOMAIN-SUFFIX,signal.org,🤖 AI Agent
  - DOMAIN-SUFFIX,whispersystems.org,🤖 AI Agent

  # === Everything else direct ===
  - MATCH,DIRECT

3. Already Have AI Routing Rules?

If you've already followed our AI Tool Proxy Routing Guide, your OpenAI/Anthropic rules are already in place. Just add the messaging platform domains above to your existing config.

24/7 Mac Mini Setup

OpenClaw's real power is continuous operation. Many users dedicate a Mac Mini as an always-on AI agent server. Here's how to keep it stable.

1. Prevent System Sleep

# Prevent sleep (recommended)
caffeinate -s -d &

# Or via System Settings:
# System Settings > Energy > disable:
#   - Put hard disks to sleep when possible
#   - Enable Power Nap
# Set "Turn display off after" to "Never"

2. Auto-start ClashX on Boot

• ClashX menu bar → Preferences → check "Launch at Login"
• Check "Set as System Proxy" to auto-enable
• If using TUN mode, ensure accessibility permissions are granted

3. Stability-Optimized DNS

dns:
  enable: true
  enhanced-mode: fake-ip
  fake-ip-range: 198.18.0.1/16
  nameserver:
    - tls://8.8.8.8:853
    - tls://1.1.1.1:853

keep-alive-interval: 30
tcp-concurrent: true

4. Health Monitoring

#!/bin/bash
# Save as ~/check-openclaw.sh, run via crontab every 5 min

if ! curl -s --proxy http://127.0.0.1:7890 \
     --connect-timeout 5 \
     https://api.openai.com/v1/models > /dev/null 2>&1; then
    osascript -e 'display notification "ClashX proxy connection issue" with title "OpenClaw Monitor"'
fi

Environment Variables & Proxy Setup

Installing OpenClaw

# Set terminal proxy (ClashX default port 7890)
export http_proxy=http://127.0.0.1:7890
export https_proxy=http://127.0.0.1:7890

# Install OpenClaw
npx openclaw init

# Verify
openclaw --version

Persistent Proxy Config

# Add to ~/.zshrc (default macOS shell):
export http_proxy=http://127.0.0.1:7890
export https_proxy=http://127.0.0.1:7890
export no_proxy=localhost,127.0.0.1,deepseek.com

# Reload
source ~/.zshrc

TUN Mode (Recommended)

Instead of setting environment variables, enable ClashX TUN mode to capture all system traffic at the network layer. Every OpenClaw request automatically goes through ClashX routing rules. Enable via: ClashX menu bar → Config → Enable TUN Mode. See TUN Mode Setup Guide for details.

Security Considerations

• Use encrypted proxy protocols — Trojan, VLESS+TLS, or Shadowsocks AEAD. API keys are transmitted through the proxy; unencrypted protocols expose them
• Avoid free/public proxies — operators can intercept your API keys and credentials
• Rotate API keys regularly — even with trusted proxies, rotate OpenAI/Anthropic keys every 1-2 months
• For sensitive data, use local models — Llama 4 Lite or Phi-4 keep everything on your machine with zero external transmission

FAQ