Setup & install
From zero to a working AI assistant in ~30 min. Walkthrough uses OpenClaw; the patterns transfer to any runtime.

Prereqs: a Mac, Linux, or cloud VM. Node.js 22+ installed. An API key from at least one LLM provider (Anthropic, OpenAI, or Google).
#1. Pick a runtime
The two open-source options most worth comparing in 2026:
OpenClaw
Single-user-first. Designed to be one person's personal assistant with multiple identities, channels, and projects. Strong on memory + cron jobs + local-first defaults. Native macOS app. What this guide uses for all examples.
- Install: one npm command
- Storage: local files in ~/.openclaw/
- Best when: it's your assistant on your machine, doing real work for you
- Docs: docs.openclaw.ai
Hermes
Multi-user-first. Designed for teams or shared deployments. More opinionated about cloud hosting, identity, and access control. Useful if you're building an assistant for an organization rather than yourself.
- Install: Docker compose / Kubernetes
- Storage: Postgres + object store
- Best when: multiple users share one assistant deployment, or org-wide policy matters
Other options worth knowing: Claude Desktop + MCP servers (slick UX, less programmable), OpenWebUI / LibreChat (self-hosted chat UI with plugin systems), Cline / Cursor / Aider(coding-focused). The setup steps that follow are OpenClaw; the concepts transfer.
#2. Install OpenClaw
# install the CLI globally via npm
npm install -g openclaw
# verify
openclaw --version
openclaw statusFirst run creates ~/.openclaw/and starts a local gateway on port 18789. Status should show the gateway as running.
#3. Add an LLM provider
OpenClaw needs at least one provider key. The easiest start is Anthropic (Claude). Generate a key at console.anthropic.comand add it via the CLI:
openclaw config providers add anthropic --key sk-ant-xxxSame pattern for OpenAI / Gemini / Ollama / others. Multiple providers can be active at once; route per-session via model overrides.
#4. Install the macOS app (optional but recommended)
For Mac users: the menu-bar app owns native permissions and exposes Mac-specific tools (screen capture, camera, system commands). Without it, the assistant still works but can't reach those surfaces.
- Download the .dmg from github.com/openclaw/openclaw/releases
- Drag OpenClaw.app to /Applications
- Launch and grant TCC permissions (Notifications, Accessibility, Screen Recording, etc.)
- Pick "Local" mode โ it'll attach to the gateway already running
#5. Set up a workspace
The workspace is where the assistant's memory + skills + project code live. Default location:~/.openclaw/workspace/.
Initialize with a quick bootstrap:
cd ~/.openclaw/workspace
touch MEMORY.md AGENTS.md USER.md SOUL.md IDENTITY.md TOOLS.md
mkdir memory skills#6. Write your first MEMORY.md
One paragraph about who you are + what you want the assistant to know. Example:
# MEMORY.md
## About me
I'm a freelance designer in Lisbon. I run a small studio
with 2 collaborators. My main clients are in tech + retail.
## Preferences
- All chats in English
- Casual but professional tone
- Push back when I'm wrong, don't be a yes-man
- Don't send anything externally without my approval
## Active projects
- ProjectX (e-commerce redesign, deadline June 15)
- Studio rebrand (Q3)This file loads at the start of every session. Edit it any time. The more the assistant knows about your context, the better its work.
#7. Run your first chat
From the terminal:
openclaw chatOr pair a messaging channel (WhatsApp, Telegram, Signal) so the assistant lives on your phone:
openclaw channels add whatsapp # follow the pairing prompt#8. Run your first cron job
Scheduled tasks are where assistants stop being chatbots and start being useful. Example: a daily 07:00 morning brief.
openclaw cron add \
--name "Morning brief" \
--cron "0 7 * * *" \
--tz "Europe/Lisbon" \
--session isolated \
--message "Give me today's calendar + weather + any urgent emails. Keep it under 15 lines."Inspect what's running:
openclaw cron list#9. Add a skill (optional)
Skills give the assistant new capabilities โ read your Notion, control a browser, fetch weather. Install via clawhub:
clawhub install notion
clawhub install weather
clawhub install browser-automation#10. Where to go from here
- Read How to prompt it โ the patterns matter more than the setup.
- Read How memory works โ to use MEMORY.md effectively.
- Read Privacy & security โ before giving the assistant real access.
- Skim the Glossary when terms feel unfamiliar.
- For full docs: docs.openclaw.ai or github.com/openclaw/openclaw.
You now have a working assistant. The work from here is briefing it well, letting it act, and iterating on what it can do. The rest of this guide is the field manual for that.