Install Openclaw Locally: Setup, Dashboard, and First Chat
Install Openclaw locally, choose one model provider, open the dashboard, and add Telegram, WhatsApp, Discord, or Slack only after first chat works.
The fastest way to get Openclaw running
If your goal is simply to try Openclaw on your own machine, the setup is much lighter than many people expect.
You do not need every channel, every integration, or a pile of API keys just to get started.
The minimum working local setup is:
- Openclaw installed on your machine
- one model provider credential
- the local Gateway running
- the browser dashboard for your first chat
That means you can get Openclaw working locally before you touch Telegram, Discord, Slack, or WhatsApp.
What you actually need
According to the official getting-started docs, the current baseline is:
- Node 24 recommended or Node 22.14+ supported
- macOS, Linux, Windows, or WSL2
- one model provider credential
The docs also note that WSL2 is more stable and recommended if you are on Windows and want the full experience.
The important part is the credential requirement. To run Openclaw in the usual way, you need one working model/auth option, not all of them.
If your search is specifically about the Windows path, use the dedicated Openclaw Windows install guide. If you are deciding whether containers are worth it, compare the Openclaw Docker setup path before you start.
In practice, the easiest choices are:
- OpenAI API key
- Anthropic API key
- OpenAI Codex sign-in if you want subscription/OAuth-style access instead of a normal API key
The official FAQ also says Openclaw can run with local-only models, but if you want the smoothest first install, starting with one hosted provider is the simpler path.
Option 1: easiest install path
The official docs recommend the installer script first.
For macOS, Linux, or WSL2:
curl -fsSL https://openclaw.ai/install.sh | bash
For Windows PowerShell:
iwr -useb https://openclaw.ai/install.ps1 | iex
That installer detects your OS, installs Node if needed, installs Openclaw, and can launch onboarding.
If you already manage Node yourself, the docs also support direct global install:
npm install -g openclaw@latest
openclaw onboard --install-daemon
If you prefer to keep everything under a local prefix such as ~/.openclaw, the official local-prefix installer is:
curl -fsSL https://openclaw.ai/install-cli.sh | bash
Step 2: run onboarding
Once Openclaw is installed, the next command is the real start of setup:
openclaw onboard --install-daemon
This guided flow walks you through:
- choosing a model provider
- adding the credential for that provider
- configuring the local Gateway
- setting up the default local runtime
The official docs describe this as the recommended way to configure a local Gateway on macOS, Linux, or Windows.
Step 3: choose the minimum credential you need
This is where a lot of first-time users overcomplicate things.
You only need one usable model/auth path to start chatting with Openclaw locally.
OpenAI API key
If you want the most straightforward hosted API setup, OpenAI is one of the simplest options.
The official Openclaw provider docs say to get your key from the OpenAI dashboard and then run:
openclaw onboard --auth-choice openai-api-key
Or, if you already exported the key:
openclaw onboard --openai-api-key "$OPENAI_API_KEY"
This is the best fit if you want a normal usage-based API workflow.
Anthropic API key
Anthropic is another common starting point. The official Openclaw Anthropic docs say new setup should use an Anthropic API key from the Anthropic Console.
During onboarding, choose the Anthropic API-key option, or run:
openclaw onboard --auth-choice anthropic-api-key
The Openclaw docs explicitly say the clearest billing path is to use an Anthropic API key instead of relying on Claude subscription auth inside Openclaw.
OpenAI Codex sign-in
If you prefer subscription-style auth instead of a normal API key, Openclaw also supports OpenAI Codex sign-in.
The official provider docs show:
openclaw onboard --auth-choice openai-codex
Or:
openclaw models auth login --provider openai-codex
This is useful if you already use Codex and want Openclaw to authenticate through that path.
Step 4: verify the local install
After onboarding, verify that the Gateway is actually running:
openclaw gateway status
The official getting-started guide says you should see the Gateway listening on port 18789.
Then open the dashboard:
openclaw dashboard
This is the fastest first chat because it does not require any messaging channel setup at all. If the dashboard loads and you can send a message, your local install is working.
If the dashboard or Gateway is the part you are debugging, use the dedicated Openclaw dashboard and Gateway guide before adding more channels.
Once the first install works, use the Openclaw update guide before treating releases, doctor checks, and Gateway restarts as separate problems.
Once the runtime is stable, the next capabilities people usually evaluate are Openclaw skills and the Openclaw browser or browser relay path.
The real minimum setup, summarized
If you want the shortest path from zero to a working local Openclaw instance, do this:
- install Openclaw
- run
openclaw onboard --install-daemon - choose one model provider auth option
- confirm
openclaw gateway status - open
openclaw dashboard
That is enough for a real local setup.
Optional: add a phone or chat channel later
Once the dashboard works, you can decide whether you want to talk to Openclaw through a messaging surface.
This is where extra credentials may be needed.
Telegram: easiest chat-channel setup
The official getting-started guide calls Telegram the fastest channel to set up because it only needs a bot token.
The official Telegram docs say the flow is:
- message @BotFather
- run /newbot
- copy the token
- store it as
TELEGRAM_BOT_TOKENor in config - start the Gateway
- approve the first pairing request
Minimal runtime commands from the docs:
openclaw gateway
openclaw pairing list telegram
openclaw pairing approve telegram <CODE>
Telegram is a very good first channel because it is simple and low-friction.
WhatsApp: no API key, but you do need a login session
WhatsApp is different. The built-in Openclaw flow does not require a separate API key, but it does require linking a WhatsApp account.
The official docs show the quick setup like this:
openclaw channels login --channel whatsapp
openclaw gateway
If you use the default pairing mode, you may also need to approve the first request:
openclaw pairing list whatsapp
openclaw pairing approve whatsapp <CODE>
So the short version is: Telegram needs a bot token, WhatsApp needs a linked session.
Discord: one bot token plus bot setup
Discord is still very manageable, but it is not as light as Telegram.
The official Discord docs say you need to:
- create a bot in the Discord Developer Portal
- copy the bot token
- enable Message Content Intent
- optionally enable Server Members Intent if you want allowlists or name lookups
- invite the bot to your server
- start the Gateway
- approve first DM pairing
Openclaw supports env fallback with:
export DISCORD_BOT_TOKEN="YOUR_BOT_TOKEN"
openclaw gateway
Discord is a good choice if you want Openclaw inside a private server or DM flow.
Slack: useful, but not the minimum path
Slack works well, but it is not the easiest first-time setup.
The official Slack docs say:
- Socket Mode requires botToken + appToken
- HTTP mode requires botToken + signingSecret
So Slack is usually something to add after your local base install already works.
If your goal is simply to get Openclaw running, start with the dashboard, Telegram, or WhatsApp before you touch Slack.
Common mistakes first-time users make
The biggest setup mistake is trying to configure everything at once.
A better sequence is:
- get the dashboard working locally
- confirm one provider credential works
- add one channel
- only then layer in more integrations
Another common problem is forgetting that the Gateway process needs access to your provider credential. The Openclaw docs repeatedly point out that env-backed keys need to be available to the process actually running the Gateway.
When local install is the right choice
Local Openclaw setup makes sense if you want:
- full control over the runtime
- a private machine-owned setup
- the simplest way to learn how Openclaw actually works
- room to customize later
It is especially good for testing, personal use, and early experimentation.
When ClawPilot is the better move
Local install is great when you want to learn and experiment. But once you move from "I want to try this" to "I want this running reliably," the operational work starts showing up.
You have to think about:
- keeping the Gateway healthy
- managing updates
- storing secrets safely
- keeping channels connected
- handling the rough edges when something breaks
That is exactly where ClawPilot fits.
If you want the Openclaw experience without owning every part of setup, hosting, and maintenance yourself, ClawPilot is the simpler path. You still get the power of Openclaw, but you skip a lot of the repetitive infrastructure and support work that comes after the first successful install.
So the practical advice is:
- install Openclaw locally if you want to learn, test, or self-manage
- use ClawPilot if you want to get to reliable usage faster
FAQ
What is the minimum API setup for Openclaw?
The minimum usual setup is one model provider credential, such as an OpenAI API key or an Anthropic API key. You do not need every provider.
Can I use Openclaw without Telegram, Discord, or Slack?
Yes. The fastest first chat is through openclaw dashboard, which does not require a messaging channel.
Which channel is easiest to add first?
According to the official docs, Telegram is the fastest channel to set up because it only needs a bot token.
Does WhatsApp require an API key?
Not in the normal built-in Openclaw setup. WhatsApp uses a linked session flow instead of a traditional API key.