How to Use OpenClaw: Complete Beginner's Guide 2026
New to OpenClaw? This step-by-step guide covers everything from installation to your first AI coding session.
What Is OpenClaw?
OpenClaw is an AI-powered coding agent that lives in your terminal. It can read your codebase, write and edit files, run commands, and help you build software faster. Unlike chatbots, OpenClaw has real access to your project files and can take actions on your behalf. It supports multiple AI models through OpenRouter, local models via Ollama, or direct API connections to providers like OpenAI and Anthropic.
How to Install OpenClaw
Installation takes under a minute. OpenClaw is available as an npm package, a standalone binary, or through Docker. The quickest way is via npm: install it globally, then run it in any project directory. OpenClaw automatically detects your project structure, reads configuration files, and adapts to your tech stack. It works on macOS, Linux, and Windows (via WSL).
See the real cost per message for 350+ models in our live comparison table.
Compare prices →Setting Up Your API Key
To use OpenClaw with cloud models, you need an API key. The easiest approach is OpenRouter, which gives you access to 350+ models through a single API key. You can also use direct API keys from OpenAI, Anthropic, or Google. Set your key as an environment variable or configure it in OpenClaw's settings. For a free start, use OpenRouter's free tier models or set up Ollama for local model inference.
Choosing the Right Model
Model selection is one of the most important decisions for your OpenClaw experience. Budget models (GPT-4.1 Nano, Gemini Flash) are fast and cheap, perfect for simple tasks. Mid-range models (Claude Sonnet 4, GPT-4.1 Mini) handle complex coding well. Premium models (Claude Opus 4.6, GPT-4.1) offer the best reasoning for architecture decisions and complex debugging. On LLM Bench, we rank every model by messages per dollar and tag the most popular ones for OpenClaw, so finding the right model for your budget takes seconds. Prices update hourly from OpenRouter.
Not sure which model fits your budget? Compare prices side by side.
Compare prices →Your First OpenClaw Session
Open your terminal in a project directory and launch OpenClaw. Start with something simple: ask it to explain a file, fix a bug, or add a feature. OpenClaw reads your code, proposes changes, and asks for confirmation before modifying files. It maintains conversation context, so you can iterate on changes across multiple messages. The more specific your instructions, the better the results.
OpenClaw Tips for Productivity
Power users get the most out of OpenClaw by: using CLAUDE.md files to set project-specific instructions, creating custom skills for repeated workflows, leveraging the /compact command to manage context length, choosing the right model for each task type (cheap for simple, premium for complex), and using OpenClaw's ability to run commands and tests to validate changes automatically.
Ready to find the best model for your budget? Compare 350+ AI models ranked by messages per dollar, updated every hour.
Open LLM Bench →