OpenClaw Configuration Guide

📘 Configuration hub: This page is your entry point for all OpenClaw configuration topics. For first-time setup, start with the Quick Start Guide and then return here to customize file structure, models, agents, and memory.

Overview

OpenClaw is a model-agnostic, self-hosted AI agent platform. Its behavior is controlled by a single main configuration file (and optional environment variables) that define the gateway, channel adapters (WhatsApp, Telegram, Discord, Slack, and 15+ more), model providers (Anthropic Claude, OpenAI, Google Gemini, DeepSeek, or local models via Ollama), agents (personalities, system prompts, skills), memory, and advanced settings. Understanding and editing this configuration lets you tailor your AI agent to your use cases while following security best practices.

This section covers:

Configuration File Guide

OpenClaw uses a single main config file (typically YAML or JSON) and a predictable directory layout for config, session data, logs, and installed skills. Knowing where files live and how sections are structured helps you edit safely, back up correctly, and troubleshoot with openclaw doctor and openclaw status.

→ Configuration File Structure & Directory Layout

Model Provider Setup

OpenClaw works with any LLM backend: Anthropic Claude, OpenAI GPT, Google Gemini, DeepSeek, and local models (e.g. via Ollama or LM Studio). You configure API keys (or local endpoints) and choose which model each agent uses. This keeps your data on your infrastructure when using local models, or lets you mix cloud and local for cost and privacy.

→ Model Providers Overview

Memory System

OpenClaw’s memory system provides persistent context across all sessions and messaging platforms. You can configure how much context is stored, where it is saved, and how it is used so your agent remembers past conversations and tasks.

→ Memory Configuration Guide

Agent Customization

You can run a single default agent or define multiple agents with different personalities, system prompts, models, and ClawHub skills. Each agent can be tied to specific channels or use cases-for example, a “support” agent for customer queries and a “personal” agent for productivity.

→ Agent Customization & Multi-Agent Setup

Advanced Settings

Advanced configuration covers performance tuning, logging, sandbox and tool restrictions, and other options for power users and production deployments. Use these after you have the basics in place.

→ Advanced Configuration

Quick Reference

Common configuration-related commands:

  • openclaw config edit - open the main config file in your default editor
  • openclaw status - show gateway and channel status
  • openclaw doctor - run diagnostics (paths, config, connectivity)
  • openclaw doctor --fix - attempt automatic fixes where supported

For credential and API key management, use environment variables or a secrets manager and follow the credential management and security best practices guides. If something breaks after a config change, see the troubleshooting guide and configuration problems section.

Prerequisites

Next Steps

Related Resources