Step 2 of 9
Planning Your Setup
Before you install OpenClaw, you need to make some decisions. These choices affect cost, privacy, performance, and what you can do with it. Let's walk through them.
💡 The Good News
You can change most of these decisions later. Don't overthink it. For beginners, we recommend: local setup + OpenAI OAuth (Codex) for best value/ease. Total cost depends on usage and plan.
Decision 1: Where Will It Run?
Local (Your Computer)
Runs on your Mac, PC, or Raspberry Pi at home.
- ✓ Free (no server costs)
- ✓ Private — your data never leaves home
- ✓ Easy setup — just install and run
- ✗ Computer must stay on
- ✗ Home internet dependent
Best for: Beginners, privacy-focused users
VPS (Cloud Server)
Runs on a rented server in the cloud (DigitalOcean, AWS, etc.).
- ✓ Always on — works 24/7
- ✓ Accessible anywhere — from any device
- ✓ Reliable — data center internet & power
- ✗ Costs $5-20/month
- ✗ More complex setup
Best for: Power users, 24/7 automation needs
Decision 2: Which AI Model?
OpenClaw works with multiple AI providers. You can even switch between them or use local models.
🎥 Provider setup walkthrough: https://www.youtube.com/watch?v=qURVA5XO84s
Claude Opus 4.6
MOST CAPABLE- • Best overall capability and reasoning (Feb 2026)
- • 1M token context window
- • Outperforms GPT-5.2 on most benchmarks
- • Premium pricing for hardest problems
Cost: ~$15-30/month for typical use
Claude Sonnet 4.6
BEST VALUE- • Matches or beats GPT-5.2 at 25-46% lower cost
- • Preferred over Opus 4.6 in 59% of comparisons
- • Excellent for coding and complex tasks
- • Sweet spot of capability vs price
Cost: ~$3-10/month for typical use
OpenAI / Codex via OAuth
BEST VALUE FOR CODEX- • Latest OpenAI flagship model
- • Strong coding and reasoning capabilities
- • Reliable and well-documented
- • GPT-5.2 mini available for cheaper tasks
Cost: ~$8-20/month for typical use
OpenAI / GPT-4o
PROVEN- • Still very capable for most tasks
- • Lower cost than GPT-5.2
- • GPT-4o mini for simple tasks
- • Good balance for beginners
Cost: ~$5-15/month for typical use
Google / Gemini 2.5 Flash
FAST & CHEAP- • Extremely fast response times
- • Very low cost per token
- • Good for high-volume tasks
- • Great for simple queries and summaries
Cost: ~$1-5/month for typical use
Moonshot / Kimi K2.5
BUDGET- • Large context window (256k tokens)
- • Often has free tiers or very cheap
- • Good balance of intelligence and cost
- • Decent for coding and complex tasks
Cost: ~$0-2/month for typical use
Local Models (Ollama)
FREE & PRIVATE- • Runs entirely on your computer — zero API costs
- • Complete privacy, no data leaves your machine
- • Requires decent GPU/CPU for good performance
- • Less capable than cloud models but improving fast
Cost: $0 (but needs good hardware)
Decision 3: What Will It Cost?
💡 How We Calculate Costs
These estimates are based on typical usage: Heartbeat check every 30 minutes (48 times daily), one morning brief per day, and about 15 chat messages daily. Your actual costs may vary depending on how much you use OpenClaw.
Typical Monthly Usage
Based on: Heartbeat every 30 min (48/day), daily morning brief, ~15 chat messages per day
Realistic Monthly AI Costs
Based on ~1.5M tokens/month usage:
Total Monthly Costs (All-In)
Web Search (Brave): $0 (2,000 free queries covers typical use)
What You Need Before Starting
Required
- ✓ Computer that stays on (Mac/PC/Linux)
- ✓ Node.js 18+ installed
- ✓ API key from an AI provider
- ✓ Basic comfort with command line
Optional (But Recommended)
- ○ Telegram account (for mobile access)
- ○ Brave Search API key (free)
- ○ Docker (for advanced memory)
Ready to Install?
You now have the context to make informed decisions. Next, we'll walk through the actual installation step by step.