Back to getting started
Agent Setup

Getting Started with OpenClaw

Run your own personal AI assistant on your devices and the channels you already use.

openclawpersonal assistantself-hostedgetting started
Verified 45 days ago

Getting Started with OpenClaw

OpenClaw is a personal AI assistant you run on your own machines. It answers you on the messaging channels and surfaces you already use — WhatsApp, Telegram, Slack, Discord, Signal, iMessage, Google Chat, web chat, and more — while keeping your data under your control.

Unlike cloud-only assistants, OpenClaw is designed to be local-first: the Gateway is the control plane, but the assistant lives in your workspace. You can run it on a laptop, a home server, a VPS, or a cloud instance, and you choose which models and tool policies it uses.

What OpenClaw is

OpenClaw is best understood as three layers:

  • Your agent — the persona, model, memory, and skills that make up the assistant.
  • The Gateway — a local Node service that routes messages, runs tools, serves the web dashboard, and connects channels.
  • Your workspace — the files, config, and persistent state that live under ~/.openclaw/workspace.

The project is open source and model-agnostic. It supports API-based providers (OpenAI, Anthropic, Google, MiniMax, Kimi, OpenRouter, etc.) and local providers such as Ollama, while giving you granular control over sandboxing, tool approval, and routing.

Strengths

  • Privacy by default. Conversations and workspace data stay on hardware you control.
  • Channel breadth. Native support for 20+ chat platforms means you can talk to the same agent from Telegram while at the store and Slack while at your desk.
  • Multi-agent ready. Built for fleets of agents with isolated workspaces, per-agent tool policies, and delegated sub-agents.
  • Skill ecosystem. Install reusable skills from ClawHub or write your own in TypeScript/JavaScript.
  • Local + cloud hybrid. Use cheap local models for routine tasks and cloud models for heavier reasoning.
  • Fail-closed security. Recent releases added stricter exec approvals, sandbox boundaries, and transcript isolation.

Requirements

Minimum practical setup:

  • Node.js 24 (recommended) or Node.js 22.19+ — the installer handles this.
  • macOS, Linux, or Windows/WSL2 — WSL2 is strongly recommended on Windows.
  • Git — used by the installer to pull the repo.
  • A model provider account or local model runtime:
    • API key route: OpenAI, Anthropic, OpenRouter, MiniMax, Kimi, Google, etc.
    • Local route: Ollama with a downloaded model.
  • Optional but useful:
    • Brave Search API key or another web-search provider for research tools.
    • A persistent host or VPS if you want always-on messaging.

Windows users can also install the native Windows Hub companion app for a GUI-first setup path.

How to get started

1. Install OpenClaw

Run the installer on macOS, Linux, or WSL2:

curl -fsSL https://openclaw.ai/install.sh | bash

On Windows PowerShell:

iwr -useb https://openclaw.ai/install.ps1 | iex

If you already manage Node yourself, you can also use npm/pnpm/bun:

npm install -g openclaw@latest
openclaw onboard --install-daemon

2. Run the onboarding wizard

openclaw onboard

The wizard walks you through:

  • Local vs remote Gateway
  • Model provider and auth (OAuth or API key)
  • Channels (Telegram, Discord, WhatsApp, Slack, etc.)
  • Workspace defaults
  • Optional background daemon

For the fastest first chat without configuring any channel, run:

openclaw dashboard

This opens the Control UI in your browser at http://127.0.0.1:18789/.

3. Choose your model

During onboarding you pick a default model. For local-first use, pair OpenClaw with Ollama. For heavier tasks, use a cloud provider. You can change this anytime:

openclaw configure
openclaw agents add <name>

4. Secure the setup

New local setups default to a strict tool profile. Review these before giving the agent broad access:

  • tools.profile — controls which tool categories are enabled.
  • sandbox.mode — controls whether group/channel sessions are sandboxed.
  • session.dmScope — controls who can message the agent directly.

Resources

Official

Community

Related SMF Clearinghouse entries