Building a Local LLM · 22/30 · Connect with an Agent
Full series and learning sequence · Practice code ZIP
OpenClaw connects a conversation interface with model providers, tools and messaging channels. It runs separately from Ollama, which handles model inference. After installation, verify the CLI, start the Gateway and test a model response before adding external channels.
Check the version and path of Node and npm prepared in Part 9. The model server uses the existing Windows Ollama.
Node Requirements and CLI Installation
Before installing, check Node and npm again. As of October 18, 2026, the official documentation requires Node 24.16 or higher or 26.1 or higher. The package’s engines are also searched, and judgment is made based on the documentation and metadata from the date of installation.
node --version
npm.cmd view openclaw version engines
npm.cmd install -g openclaw@latest
openclaw --version
openclaw --helpIf your PowerShell policy blocks openclaw.ps1, you can run the same command with openclaw.cmd. Please check before relaxing the entire execution policy to administrator privileges. If Node is in multiple paths, the where.exe results will first look for the actual installation in use.
Gateway and model server
Gateway coordinates channels, sessions, and provider connections. Inference is handled by the Ollama server or external API. If the gateway is turned on but the model server is turned off, you will not receive an answer, so check both states separately.
| configuration | Location of this series |
|---|---|
| Node·OpenClaw CLI | Windows standard user experience |
| Gateway·Dashboard | Same PC, check basic 18789 instructions |
| model server | Part 2 Windows Ollama 11434 |
| external channel | No connection in first lab |
The first check is done on your local dashboard. If you connect Messenger first, there are more items to check, including bot tokens, user whitelists, and external communications. After receiving the model’s answers on the dashboard, add channels one by one.
Local provider onboarding
openclaw onboard --classicRead the onboarding instructions and select Ollama and Local only. Make sure you don’t confuse this with existing logins or routes using external providers. Address and model will be checked in the next article. If the on-screen menu has changed, follow the current official onboarding instructions.
If you have already used settings, back up the settings and workspace first. Your provider or default model may change during the onboarding process. Start with a separate profile or lab account and do not designate sensitive work folders as workspaces.
Check resident services and responses
If you intend to continue running the Gateway after normal conversation is confirmed, apply the command below. On native Windows, scheduled tasks or guided login initiation methods can be used. Since WSL is a Linux user service, the behavior and startup conditions are different.
openclaw gateway install
openclaw gateway status
openclaw dashboard
# 실습 종료 시
openclaw gateway stopAfter the dashboard opens, check to see if you get real answers to your short questions. When sharing logs, we erase the key and original text and do not disclose the entire settings. We will continue to check Ollama addresses and tool calls in the next article.
reference material
Documentation and model tags checked: October 18, 2026. Primary lab environment: Windows 11 x64.
.jpg){kind=link}