Using OpenClaw on Windows

github.com

GitHub - openclaw/openclaw: Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞

Your own personal AI assistant. Any OS. Any Platform. The lobster way. 🦞 - openclaw/openclaw

Overall Flow

Documentation as of March 14, 2026

OpenClaw (formerly ClaudeBot/MoltBot) is an open-source AI Agent that runs on your local machine. When you send commands through messengers such as Telegram, WhatsApp, or Discord, it carries out tasks on your behalf, including file manipulation, sending emails, executing code, and managing schedules.

Before We Begin

For most paths, we will use the recommended installation option, which installs everything immediately during setup.

<UserName> refers to the username (path) where the application is typically installed.

1. Prerequisites

Step 1: Install WSL2 and Ubuntu 24.04

Open PowerShell as Administrator, then enter the following command.

wsl --update
wsl --install -d Ubuntu-24.04

After installation is complete, reboot the system. Once the Ubuntu terminal opens after rebooting, set your username and password.

If it does not open automatically:

wsl -d Ubuntu-24.04

Troubleshooting: WSL_E_DISTRO_NOT_FOUND Error

This error can occur during the initial activation of the WSL feature.

wsl --update
wsl --list --online
wsl --install -d Ubuntu-24.04

Step 2: Enable systemd

Run the following in the Ubuntu terminal.

sudo tee /etc/wsl.conf >/dev/null <<'EOF'
[boot]
systemd=true
EOF

Shut down WSL from PowerShell, then restart it.

wsl --shutdown

1.1 Install Node.js 22 or Later

OpenClaw requires Node.js 22 or later.

If you do not have it yet, download the LTS version from the official site.

• Download: https://nodejs.org

• Select the "Recommended For Most Users" version

• Proceed with the defaults in the installation wizard

Verify the installation:

node --version
# v22.x.x or higher is valid

PS C:\Users\Pc> node --version v22.17.0 PS C:\Users\Pc>

1.2 Installing Git (recommended)

Some npm packages require Git when downloading source from GitHub.

• Download: https://git-scm.com

• Proceed with the defaults in the installation wizard

1.3 CMD vs. PowerShell

When opening a terminal on Windows, CMD (Command Prompt) and PowerShell are different environments. The OpenClaw installation script is PowerShell-only, so you must run it in PowerShell.

How to tell them apart:

• CMD prompt: C:\Users\<UserName>>

• PowerShell prompt: PS C:\Users\<UserName>>

To switch from CMD to PowerShell, type powershell.

1.4 Checking the PowerShell Execution Policy

Windows' default execution policy may block script execution. Open PowerShell in Administrator mode and change the execution policy:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

When a confirmation prompt appears, type Y.

2. Installing OpenClaw

Open PowerShell in Administrator mode and run the installation script:

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

What this script does:

• - Check the Node.js version

• - Globally install the OpenClaw CLI via npm

• - Create the initial configuration directory (~/.openclaw/)

• - Run the onboarding wizard

3. Configuring PATH (Critical)

Even after installing OpenClaw, the openclaw command is often not recognized. This is because the npm global binary path is not registered in the user's PATH environment variable.

3.1 Diagnosing the Problem

C:\Users\<UserName>>openclaw --version
'openclaw' is not recognized as an internal or external command, operable program or batch file.

3.2 Verifying the Binary Exists

First, confirm that the installation itself succeeded:

npm prefix -g
# Example output: C:\Users\<UserName>\AppData\Roaming\npm

dir "C:\Users\<UserName>\AppData\Roaming\npm\openclaw*"
# If openclaw, openclaw.cmd, and openclaw.ps1 exist, the installation is complete

Verify it works by specifying the path directly:

"C:\Users\<UserName>\AppData\Roaming\npm\openclaw.cmd" --version
# If the version is printed, the binary is working correctly

3.3 Methods for Adding to PATH (2 Options)

Method A: GUI (sysdm.cpl) - Safest

1. Press Win + R, type sysdm.cpl, and press Enter

2. Click the Advanced tab

3. Click the Environment Variables button

4. In the User variables section at the top, select Path and click Edit

5. Click New

6. Enter C:\Users\<UserName>\AppData\Roaming\npm

7. OK -> OK -> OK

8. You must fully close and reopen any CMD/PowerShell windows for the changes to take effect

Method B: PowerShell - Safely append to the existing value

$current = [Environment]::GetEnvironmentVariable("PATH", "User")
[Environment]::SetEnvironmentVariable("PATH", "$current;C:\Users\<UserName>\AppData\Roaming\npm", "User")

Warning: Never use setx PATH in CMD.

The setx PATH "%PATH%;new-path" command merges the user PATH and the system PATH, then overwrites the user PATH with the result, and the content gets truncated due to the 1024-character limit. This causes previously registered paths for Ollama, Go, Python, and others to be lost.

A real-world example:

CodeCopy

C:\Windows\System32>setx PATH "%PATH%;C:\Users\<UserName>\AppData\Roaming\npm"
WARNING: The data being saved is truncated to 1024 characters.
SUCCESS: The specified value has been saved.

After this, the ollama command will no longer be recognized. See Section 4 for detailed recovery steps.

3.4 Verification

In a new terminal:

openclaw --version
ollama --version

If both work correctly, the PATH configuration is complete.

4. Recovering from PATH Loss (when a setx accident occurs)

This section covers how to recover if you accidentally ran setx PATH and wiped out existing paths.

4.1 The Mechanism by Which setx Destroys PATH

There are two types of PATH environment variables in Windows:

Final PATH = System PATH + User PATH

When you reference %PATH% in CMD, you get this combined value. When you run setx PATH "%PATH%;새경로":

1. It attempts to write the entire combined PATH (thousands of characters) into the user PATH

2. The contents of the system PATH get duplicated and copied into the user PATH

3. The trailing portion is truncated due to the 1,024-character limit

4. Paths at the end, such as Ollama and Rancher Desktop, are lost

4.2 Diagnosing the Current State

In PowerShell, check the user PATH and system PATH separately:

Write-Host "=== USER PATH ==="
[Environment]::GetEnvironmentVariable("PATH", "User")
Write-Host ""
Write-Host "=== SYSTEM PATH ==="
[Environment]::GetEnvironmentVariable("PATH", "Machine")

setx only overwrites the user PATH. The system PATH is likely intact. The probable state is that the entire system PATH has been duplicated into the user PATH, and the trailing portion (including Ollama, etc.) has been lost due to the 1,024-character truncation.

4.3 Finding the Lost Paths

Find the actual locations of programs that are not recognized, such as Ollama:

where /R "C:\Users\<UserName>" ollama.exe
# Example output: C:\Users\<UserName>\AppData\Local\Programs\Ollama\ollama.exe

4.4 Resetting the User PATH

Since the system PATH already includes Git, Python, Go, Node, Windows default paths, and so on, the user PATH only needs to contain the paths for programs the user installed directly.

In PowerShell:

[Environment]::SetEnvironmentVariable("PATH", "C:\Users\<UserName>\AppData\Local\Programs\Ollama;C:\Users\<UserName>\AppData\Roaming\npm", "User")

Add any other user program paths as needed, separated by semicolons.

4.5 Verification

Fully close and reopen CMD/PowerShell, then run:

ollama --version
openclaw --version

If both are recognized, the recovery is complete.

4.6 Lessons Learned

Always add to PATH using the GUI or PowerShell. Do not use setx PATH.

5. Onboarding Wizard

The installation script starts onboarding automatically. If you skipped it, run it manually:

openclaw onboard --install-daemon

Items configured in the wizard:

Press Yes to proceed.

After this, a list of model names will appear, which will vary depending on which AI you are using.

Since API costs tend to be higher than expected, the personal recommendation is to integrate with an OpenAI subscription plan or use a local LLM instead.

What is Ollama?

Ollama is an open-source platform that makes it easy to run and use large language models (LLMs) on your local computer. It lets you run models like DeepSeek or LLaMA with simple commands and use them like an API, without any complex configuration.

6. Ollama Integration

Starting from Ollama version 0.17, you can integrate it with OpenClaw. This section assumes that Ollama is already installed on Windows.

6.1 The ollama launch Command

Since the OpenClaw installation and PATH configuration are complete, it should now work correctly:

ollama launch openclaw

Here you select a model; for this guide, we will choose the gpt-oss 20B model.

Example output on successful startup:

This will modify your OpenClaw configuration:
  C:\Users\<Username>\.openclaw\openclaw.json
Backups will be saved to C:\Users\<Username>\AppData\Local\Temp\ollama-backups/
Added gpt-oss:20b to OpenClaw
Launching OpenClaw with gpt-oss:20b...

Tasks handled by Ollama:

• Back up existing configuration

• Display the model selection screen

• Automatically register the selected model in the OpenClaw configuration

• Start the gateway

• Launch the TUI (Text User Interface) console

• 6.2 Model Selection Criteria

  OpenClaw recommends a minimum context window of 64K tokens.

  Ollama Cloud Models (recommended, free tier available):

  Model	Features
  gpt-oss:20b	General-purpose, default choice
  Kimi-K2.5	Cost-efficient

  Model

  gpt-oss:20b

  Features

  General-purpose, default choice

  Model

  Kimi-K2.5

  Features

  Cost-efficient

  Local Models (choose based on VRAM):

  Model	VRAM	Use Case
  qwen3:8b	8GB	Lightweight Q&A
  qwen2.5-coder:32b	24GB	Coding-specialized, tool calling support
  deepseek-r1:32b	24GB	Reasoning-specialized
  glm-4.7-flash	25GB	Reasoning + code generation

  Model

  qwen3:8b

  VRAM

  8GB

  Use Case

  Lightweight Q&A

  Model

  qwen2.5-coder:32b

  VRAM

  24GB

  Use Case

  Coding-specialized, tool calling support

  Model

  deepseek-r1:32b

  VRAM

  24GB

  Use Case

  Reasoning-specialized

  Model

  glm-4.7-flash

  VRAM

  25GB

  Use Case

  Reasoning + code generation

  6.3 If you want to change settings only without starting the gateway

  CodeCopy

  ollama launch openclaw --config
  

  6.4 Manually changing the model

  To change the model while the gateway is already running:

  PowerShellCopy

  openclaw models set ollama/qwen2.5-coder:32b
  

  Or edit the configuration file directly:

  PowerShellCopy

  notepad %USERPROFILE%\.openclaw\openclaw.json
  

  Configuration example (the models section in openclaw.json):

  JSONCopy

  {
    "models": {
      "providers": {
        "ollama": {
          "baseUrl": "http://localhost:11434/v1",
          "apiKey": "ollama-local",
          "api": "openai-completions",
          "models": [
            {
              "id": "gpt-oss:20b",
              "name": "gpt-oss:20b",
              "reasoning": false,
              "input": ["text"],
              "cost": {
                "input": 0,
                "output": 0,
                "cacheRead": 0,
                "cacheWrite": 0
              },
              "contextWindow": 131072,
              "maxTokens": 8192
            }
          ]
        }
      }
    },
    "agents": {
      "defaults": {
        "model": {
          "primary": "ollama/gpt-oss:20b"
        }
      }
    }
  }
  

  Restart the gateway after modifying the configuration file:

  PowerShellCopy

  openclaw gateway restart

  

Afterward, verify the settings through the web dashboard.

The dashboard is accessible by default at http://127.0.0.1:18789/.

7. GPT Integration via CLI

Clicking OPENAI and completing OAuth lets you link your GPT subscription, or you can issue an API key from OpenAI and access it through the API.

8. Messenger Channel Integration (Telegram)

Telegram has the lowest setup difficulty.

8.1 Creating a Telegram Bot

1. Search for @BotFather in the Telegram app and start a conversation.

2. Enter the /newbot command.

3. Enter a bot name (e.g., MyOpenClaw).

4. Enter a bot username (e.g., my_openclaw_bot; it must end with _bot).

5. Copy the Bot Token issued by BotFather (format: 123456789:ABCdefGHI...).

8.2 Registering the Telegram Token in OpenClaw

If you selected Telegram in the onboarding wizard, a token input screen will appear.

openclaw onboard

After registering the model and selecting the Telegram channel,

enter the token in the token input field.

To register manually:

openclaw channels add --channel telegram --token <YOUR_BOT_TOKEN>

Note: You can also register via openclaw configure --section channels (the interactive wizard), but when registering directly through the CLI, you must use the channels add command. openclaw config set channels.telegram.token ... does not work (the key name is different).

After registering the token, you must restart the gateway for the changes to take effect:

openclaw gateway restart

8.2.1 401 Unauthorized Error

If the following error repeats in the gateway log:

[telegram] telegram deleteWebhook failed: Call to 'deleteWebhook' failed! (401: Unauthorized)
[telegram] [default] channel exited: Call to 'deleteWebhook' failed! (401: Unauthorized)
[telegram] [default] auto-restart attempt 1/10 in 5s

Cause: The Telegram bot token is incorrect or has expired.

Resolution:

1. In Telegram, go to @BotFather -> /mybots -> select your bot -> API Token to reconfirm the token.

2. If the token is incorrect, reissue it with /revoke.

3. Register again with the new token:

openclaw channels add --channel telegram --token <NEW_TOKEN>
openclaw gateway restart

If the token issue persists, try creating a new bot.

8.3 Pairing

1. Find the bot you just created in Telegram and send /start.

2. The bot sends a message containing the pairing code.

3. Approve the pairing in the terminal:

openclaw pairing approve telegram <CODE>

8.4 Verification

Send any message to the bot in Telegram.

Once you receive a response, the integration is complete.

9. Messenger Channel Integration (Discord)

You can integrate Discord in addition to, or instead of, Telegram.

9.1 Creating a Discord Bot

1. Go to https://discord.com/developers/applications

2. Click New Application and enter a name (e.g., OpenClawBot).

3. Click the Bot tab in the left menu.

4. Click Reset Token to copy the bot token.

9.2 Configuring Privileged Gateway Intents (Required)

In the Bot tab, scroll down to find the Privileged Gateway Intents section. Enable all three of the following:

• Presence Intent - Detects online status

• Server Members Intent - Reads server member list

• Message Content Intent - Reads message content (required)

After enabling them, be sure to click Save Changes.

If you skip this step, the bot will not work. The following error will repeat in the gateway logs:

PowerShellCopy

[discord] discord: gateway closed with code 4014 (missing privileged gateway intents)
[discord] [default] auto-restart attempt 1/10 in 5s

If you see this error, check that the Intent toggles are enabled in the Discord Developer Portal. Once you enable the toggles and save, the gateway will connect successfully on its next automatic retry.

9.3 Inviting the Bot to a Server

1. Click OAuth2 in the left menu.

2. Scroll down to find the OAuth2 URL Generator section.

3. Check bot in the Scopes list (5th item from the top in the left column).

4. Checking bot will make a Bot Permissions panel appear below.

5. Check the following under Bot Permissions:

   • Send Messages

   • Read Message History

   • Read Messages/View Channels

   • Embed Links

   • Attach Files

6. The Generated URL will appear at the bottom of the page.

7. Copy the URL and paste it into your browser.

8. Select the server you want to invite the bot to and confirm.

9.4 Registering the Discord Token in OpenClaw

openclaw channels add --channel discord --token "YOUR_BOT_TOKEN"

Note: In PowerShell, the token must be wrapped in quotes. Special characters (such as .) included in a Discord bot token may be interpreted as shell operators. In CMD, it works without quotes, but wrapping it in quotes as a habit is recommended.

After registering, restart the gateway:

openclaw gateway restart

9.5 Pairing

Send any message to the bot via DM in Discord, and a pairing code will appear.

openclaw pairing approve discord <CODE>

9.6 Verification

Send a message to the bot via DM in Discord. If you receive a response, the integration is complete.

Note: Test with a DM first, not a server channel. To use the bot in a server channel, either call it with a mention (@OpenClawBot) or adjust the mention gating setting in the OpenClaw configuration.

10. Starting and Verifying the Gateway

# Start gateway
openclaw gateway start

# Check status
openclaw gateway status

# Open web dashboard
openclaw dashboard

The web dashboard is accessible by default at http://127.0.0.1:18789/, and it provides a convenient GUI overview of the status at a glance.

To change the port:

openclaw gateway --port 18790

11. Workspace Customization

After installation, the following structure is created in the %USERPROFILE%\.openclaw\ directory:

%USERPROFILE%\.openclaw\
  openclaw.json          # Main configuration (models, channels, gateway, etc.)
  workspace\
    AGENTS.md            # Agent behavior guidelines (bootstrapper)
    SOUL.md              # Persona (personality, tone, attitude)
    USER.md              # User information (name, work context, preferences)
    MEMORY.md            # Long-term memory (learned from conversations)
    IDENTITY.md          # Agent self-introduction (name, emoji, etc.)
    skills/              # Skill plugin directory

11.1 Role of Each File

OpenClaw reads the workspace files at the start of each session to build the agent's personality and context. Because the agent has no memory between conversations, these files effectively serve as its memory.

11.2 Differences Between SOUL.md and AGENTS.md

The differences between these two easily confused files:

SOUL.md = "Who it is" (personality)

• Speech style, attitude, and humor style

• Level of sarcasm, standards for giving compliments

• Emotional boundaries (when to be serious)

AGENTS.md = "What to do and how" (behavioral rules)

• Tool usage permissions and restrictions

• Task execution procedures and priorities

• Response format (code style, language, etc.)

• List of actions requiring approval

For example, "Reply with a sarcastic tone" goes in SOUL.md, while "Always get user confirmation before sending an email" goes in AGENTS.md.

11.3 Editing SOUL.md

Configure the agent's persona:

notepad %USERPROFILE%\.openclaw\workspace\SOUL.md

11.4 Editing USER.md

Enter user information so the agent can use it as context:

notepad %USERPROFILE%\.openclaw\workspace\USER.md

11.5 Editing AGENTS.md

Define the agent's behavioral rules:

notepad %USERPROFILE%\.openclaw\workspace\AGENTS.md

11.6 Changing the Model Within a Session

You can switch models even during an ongoing conversation:

/model opus4.5

12. Adding Skills

Skills are plugins that add functionality to OpenClaw.

12.1 Searching and Installing Skills with the ClawHub CLI

npm i -g clawhub

clawhub search "calendar"
clawhub install <skill-slug>
clawhub list

12.2 Creating a Skill File Manually

You can also create a SKILL.md file directly in the workspace/skills/ folder. It will be loaded automatically when a new session starts.

13. Key Commands Reference

14. Troubleshooting

14.1 The openclaw command is not recognized

Cause: The npm global path is not in PATH.

Diagnosis:

# Check binary location
npm prefix -g
# Example Output: C:\Users\<UserName>\AppData\Roaming\npm

# Check if files exist
dir "C:\Users\<UserName>\AppData\Roaming\npm\openclaw*"

# Test execution using direct path
"C:\Users\<UserName>\AppData\Roaming\npm\openclaw.cmd" --version

If the version is printed when using the direct path, the installation is complete and you only need to add the path to PATH. See Section 3.3 (How to add PATH).

14.2 The ollama command is not recognized

Cause A: User PATH lost due to setx PATH (see Section 12.3 below)

Cause B: Ollama path is not registered in PATH

Diagnosis:

where /R "C:\Users\<UserName>" ollama.exe
# Example output: C:\Users\<UserName>\AppData\Local\Programs\Ollama\ollama.exe

If the path is printed, Ollama is installed but not in PATH. Add that path's directory to PATH using the method described in Section 3.3.

14.3 Environment Variable Loss Caused by setx PATH

Symptoms:

• After running setx PATH "%PATH%;new-path" in CMD,

• the message "Warning: The data being saved is truncated to 1024 characters" is displayed,

• and commands that previously worked, such as ollama, go, and python, are no longer recognized.

Root Cause Analysis:

%PATH% is the combined value of the user PATH and the system PATH. setx PATH overwrites only the user PATH. The combined long string is truncated at 1024 characters and written to the user PATH, causing paths near the end (such as Ollama) to be lost.

Recovery: See Section 4 (PATH Loss Recovery).

Prevention: Always add to PATH using the GUI (sysdm.cpl) or PowerShell.

14.4 Errors When Running PowerShell-Only Commands in CMD

Symptoms:

C:\Windows\System32>[Environment]::SetEnvironmentVariable(...)
The filename, directory name, or volume label syntax is incorrect.

Cause: [Environment]::SetEnvironmentVariable() is a PowerShell-only .NET syntax. It cannot be executed in CMD (Command Prompt).

Fix:

• Type powershell in CMD to switch to PowerShell, then run the command

• Or open PowerShell directly from the Start menu

How to tell the difference:

• CMD prompt: C:\Users\<UserName>>

• PowerShell prompt: PS C:\Users\<UserName>>

14.5 Command Still Not Recognized After Changing PATH

Cause: Changes to environment variables are not applied to terminals that are already open.

Fix: You must completely close and reopen the CMD/PowerShell window for the changes to take effect. No matter how many times you re-enter a command in the same window, it will keep using the old PATH.

14.6 The ollama Command Exits with No Output (system32 Ghost File)

Symptoms: Running ollama --version in PowerShell returns to the prompt with no output. Or, running ollama launch openclaw causes an "Choose an app to open 'ollama'" dialog to appear.

Diagnosis:

Get-Command ollama
# If Source is C:\Windows\system32\ollama, this is an issue

(Get-Item "C:\Windows\system32\ollama").Length
# If 0, it confirms a ghost file

Cause: A 0-byte empty ollama file exists at C:\Windows\system32. This file takes priority over the real ollama.exe (in the user PATH), intercepting calls to it. It may have been created during the WSL installation process.

Fix: In an administrator PowerShell:

Remove-Item "C:\Windows\system32\ollama" -Force

After deletion, in a new terminal:

ollama --version
# If the version prints correctly, the issue is resolved

14.7 ollama launch openclaw not supported (version issue)

Symptom: Running ollama launch openclaw produces no response or throws an error.

Diagnosis:

ollama --version

Cause: ollama launch openclaw is supported in Ollama 0.17 or later. This command does not exist in version 0.16.x or below.

Fix: Download and install the latest version from https://ollama.com/download/windows.

After installation, verify:

ollama --version
# If Source is C:\Windows\system32\ollama, this is an issue

14.8 node.exe crash (0xc0000005)

This occurs when Windows Defender incorrectly flags node.exe as a threat.

Fix:

1. Temporarily disable Windows Defender real-time protection

2. Re-run the installation script

3. Re-enable real-time protection after installation completes

14.9 Script execution blocked in PowerShell

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

14.10 "No" is automatically selected in the onboarding security prompt

Symptom: After running ollama launch openclaw, "No" is automatically selected in the onboarding security prompt without any key input, and the process exits.

o  I understand this is personal-by-default and shared/multi-user use requires lock-down. Continue?
|  No
Error: exit status 1

Cause: In native Windows CMD, the interactive prompt of the OpenClaw TUI does not recognize key input. OpenClaw itself warns:

Windows detected -- OpenClaw runs great on WSL2!
Native Windows might be trickier.

Fix: Run it in Windows Terminal. Search for "Windows Terminal" in the Microsoft Store, install it, and run OpenClaw from within it; TUI input will then work correctly.

14.11 Model connection failure in ollama launch openclaw

Check whether the Ollama service is running:

ollama ps

If it is not running, start the Ollama app, or:

ollama serve

14.12 Gateway fails to start

# Diagnose
openclaw doctor

# Check logs
openclaw gateway --verbose

14.13 Out of memory during npm install

In PowerShell:

$env:NODE_OPTIONS="--max-old-space-size=4096"
iwr -useb https://openclaw.ai/install.ps1 | iex

In CMD:

set NODE_OPTIONS=--max-old-space-size=4096

14.14 Discord bot connection failure (code 4014)

Symptom: The following repeats in the gateway log:

[discord] discord: gateway closed with code 4014 (missing privileged gateway intents)
[discord] [default] auto-restart attempt 1/10 in 5s

Cause: Privileged Gateway Intents are disabled in the Discord Developer Portal.

Fix:

1. Go to https://discord.com/developers/applications

2. Click your bot app -> left Bot tab

3. Scroll down and turn ON all three options under the Privileged Gateway Intents section:

   • Presence Intent

   • Server Members Intent

   • Message Content Intent

4. Click Save Changes

5. The gateway retries automatically, so no manual restart is needed. If the logs show a successful connection with no errors, you're all set.

14.15 Error when entering a Discord token in PowerShell

Symptom:

openclaw channels add --channel discord --token <MTQ3NzY3...>
At line:1 char:49
+ ... --token <MTQ3NzY3...
+             ~
The '<' operator is reserved for future use.

Cause: PowerShell interprets <> or other special characters in the token as shell operators.

Fix: Wrap the token in quotes:

openclaw channels add --channel discord --token "MTQ3NzY3..."

Or use the interactive wizard, which avoids quoting issues:

openclaw configure --section channels

15. Security Considerations

The creator of OpenClaw (Peter Steinberger) explicitly warns in the official documentation:

"Running AI agents on a personal computer is dangerous."

Principles to follow:

1. Keep it isolated from environments with financial information. Do not run it under a user account that has banking apps, brokerage accounts, or cryptocurrency wallets.

3. It is recommended to create a separate user account for running it. Create an additional standard user account in Windows and operate it in isolation.

5. Do not install skills from unknown sources. Attack cases involving malicious skills have been reported.

7. Exercise extra caution when using local models. Smaller local models can misinterpret intent and perform incorrect file operations. Agent actions such as file manipulation or automation are safer to run with API models (Claude, GPT, etc.).

9. Run security audits periodically.

   PowerShellCopy

   openclaw doctor

   16. Cost

   OpenClaw itself is open-source and free.

   However, costs may arise from LLM usage:

   Category	Cost	Billing method
   Ollama local model	Free (only electricity costs apply)	None
   Ollama Cloud Models	Free tier available; charges apply when exceeded	Pay-as-you-go
   ChatGPT Plus (Codex CLI)	$20/month flat rate	Unlimited
   ChatGPT Pro	$200/month flat rate	Unlimited (higher-tier models available)
   Anthropic Claude API	Charged based on usage	Pay-as-you-go
   OpenAI GPT API	Charged based on usage	Pay-as-you-go

   Category

   Ollama local model

   Cost

   Free (only electricity costs apply)

   Billing method

   None

   Category

   Ollama Cloud Models

   Cost

   Free tier available; charges apply when exceeded

   Billing method

   Pay-as-you-go

   Category

   ChatGPT Plus (Codex CLI)

   Cost

   $20/month flat rate

   Billing method

   Unlimited

   Category

   ChatGPT Pro

   Cost

   $200/month flat rate

   Billing method

   Unlimited (higher-tier models available)

   Category

   Anthropic Claude API

   Cost

   Charged based on usage

   Billing method

   Pay-as-you-go

   Category

   OpenAI GPT API

   Cost

   Charged based on usage

   Billing method

   Pay-as-you-go

   When an agent operates autonomously, token consumption is hard to predict. If you use a pay-as-you-go API, it is recommended to monitor usage and set daily limits.

17. Closing Thoughts

If you have followed along this far, OpenClaw should be running properly. Send a message via Telegram and it will respond. I set mine up in a Docker environment rather than Windows, but if you already know how to configure Docker, you can figure it out on your own without reading this guide.

That said, to be honest, I have my doubts about how useful this actually is.

The concept of "AI controlling my computer on my behalf" sounds impressive. But in practice, most tasks are simply faster to do yourself. Moving a single file by typing a command in Telegram and waiting for confirmation takes longer than just dragging it in File Explorer by about three seconds.

So why does this ecosystem exist at all?

The answer is simple: fear-based business.

It sells the anxiety that "if you can't build an AI Agent, you'll fall behind." It makes you feel like an inefficient person if you don't have an AI assistant on your computer. The more complex the installation process, the more troubleshooting involved, the greater the sense of accomplishment when you finish. Even if that sense of accomplishment has nothing to do with actual productivity gains.

I personally call this false accomplishment. It is not that any productivity has increased; it simply sells the feeling of having solved something. In my case the installation is fairly straightforward, but for non-developers it is quite a difficult process.

In fact, the longest section in this guide is the troubleshooting section. What does that tell you?

There are only a handful of moments when a local AI Agent is genuinely meaningful: when the repetitive task is clearly defined, that task actually occurs frequently, and delegating it is more advantageous than doing it yourself. If those conditions are not met, it is just a power-hungry chatbot. What advantage does it have over simply chatting on a chatbot website?

With Python, simple Excel automation tools are fairly easy to build.

Cross-section context issues mean Excel tool modifications often turn out to be less impressive than expected.

These days, PowerPoint tasks are mostly handled well too. If you open a CLI and spin up a multi-agent setup for coding, there is no particular reason to use OpenClaw for that either.

When you carefully examine the act of automating with agents, most things do not actually need to be automated. That said, there will certainly be cases where it proves useful for some people.

Having more options is genuinely great. That said, I remain skeptical about whether it actually delivers the explosive productivity gains that the media claims it does.