OpenClaw AI Assistant: Build Your Personal AI Agent (2026)

What is an OpenClaw AI Assistant?

An OpenClaw AI assistant is a personal AI agent that you define, deploy, and own. Unlike cloud-based AI tools where you interact through a web interface and your data passes through third-party servers, an OpenClaw assistant runs locally on your machine or your own server. You control what it knows, what it can access, and where it operates.

At its core, every OpenClaw assistant is defined by a single file called SOUL.md. This markdown file describes the assistant's identity, personality, rules, and skills. The OpenClaw gateway reads this file and turns it into a working AI agent that can respond to messages, execute tasks, and interact with external services.

What makes an OpenClaw assistant different from a simple LLM wrapper is the skills system. Skills give your assistant the ability to browse the web, read files, query databases, call APIs, and connect to analytics platforms. Combined with built-in chat platform integrations for Telegram, Slack, and Discord, your assistant becomes a fully operational tool that lives where you already work.

What Can Your AI Assistant Do?

The capabilities of your assistant depend entirely on what skills you enable and what data sources you connect. Here are the most common use cases people build with OpenClaw assistants:

Setting Up Your First Assistant

Every OpenClaw assistant starts with a SOUL.md file. This file is the single source of truth for your assistant's behavior. Let us walk through building one from scratch.

The SOUL.md file has four main sections: Identity, Personality, Rules, and Skills. Identity tells the agent who it is. Personality shapes how it communicates. Rules define boundaries and guidelines. Skills give it capabilities beyond basic conversation.

# BI Assistant

## Identity
- Name: Atlas
- Role: Business Intelligence Analyst
- Model: claude-sonnet-4-20250514

## Personality
- Data-driven and precise
- Explains metrics in plain language
- Proactively highlights anomalies and trends
- Keeps responses concise unless asked for detail

## Rules
- Always include the date range in reports
- Compare metrics to the previous period by default
- Flag any metric that changed more than 20% week-over-week
- Never fabricate data. If a query fails, say so clearly
- Respond in English only

## Skills
- browser: Research industry benchmarks
- file_manager: Read and write local report files

## Channels
- telegram: true

Once your SOUL.md is ready, registering and starting the assistant takes two commands:

# Register the agent with the OpenClaw CLI
openclaw agents add atlas --workspace ./agents/atlas

# Start the gateway (serves all registered agents)
openclaw gateway start

Your assistant is now running. If you enabled Telegram in the Channels section, you can open Telegram and start messaging it immediately. Every change you make to the SOUL.md file takes effect when you restart the gateway.

Connecting to Data Sources

A personal AI assistant becomes truly useful when it can access your actual data. OpenClaw supports connecting to analytics platforms, payment systems, and databases through its skills and environment configuration.

Google Analytics 4

Connect your assistant to GA4 to query traffic data, user behavior, and conversion metrics. You need a GA4 property ID and a service account with read access. Add the credentials to your environment and reference the GA4 skill in your SOUL.md.

# .env file in your agent workspace
GA4_PROPERTY_ID=372869399
GOOGLE_SERVICE_ACCOUNT_KEY=./service-account.json

Mixpanel

Mixpanel gives your assistant access to event-level analytics, funnel analysis, and user segmentation. With a Mixpanel project token and API secret, your assistant can pull funnel conversion rates, event counts, and user property distributions on demand.

MIXPANEL_TOKEN=your_project_token
MIXPANEL_API_SECRET=your_api_secret

Stripe

Connect Stripe to let your assistant report on revenue, subscriptions, churn, and payment activity. It can pull daily revenue summaries, list recent charges, check subscription counts, and flag failed payments. Use a restricted API key with read-only permissions for security.

STRIPE_SECRET_KEY=sk_live_your_restricted_key

PostgreSQL and other databases

For custom data, connect your assistant directly to a PostgreSQL, MySQL, or SQLite database. The assistant can run read-only queries to answer questions about your application data, user tables, or any custom metrics you track. Always use a read-only database user to prevent accidental modifications.

DATABASE_URL=postgresql://readonly_user:password@host:5432/mydb

You can connect all of these simultaneously. A single assistant with GA4, Mixpanel, Stripe, and database access becomes a unified analytics layer that you interact with through natural language instead of four separate dashboards.

Connecting to Chat Platforms

One of OpenClaw's strongest features is built-in support for chat platforms. Your assistant is not trapped in a terminal. It lives where you already communicate.

Telegram

Telegram is the most popular channel for OpenClaw assistants. Create a bot through BotFather, grab the token, and add it to your environment. Enable the Telegram channel in your SOUL.md, and your assistant is accessible from your phone, desktop, and web.

## Channels
- telegram: true

# In your .env file:
# TELEGRAM_BOT_TOKEN=your_bot_token_from_botfather

Slack

For team environments, Slack integration lets your assistant join channels, respond to direct messages, and participate in threads. Create a Slack app, configure the bot permissions, and add the token. Your assistant appears as a regular team member in your workspace.

## Channels
- slack: true

# In your .env file:
# SLACK_BOT_TOKEN=xoxb-your-slack-bot-token
# SLACK_APP_TOKEN=xapp-your-slack-app-token

Discord

Discord works similarly. Create a bot in the Discord Developer Portal, invite it to your server, and add the token. Your assistant can respond in channels, handle direct messages, and work alongside your community or team.

## Channels
- discord: true

# In your .env file:
# DISCORD_BOT_TOKEN=your_discord_bot_token

You can enable multiple channels at once. The same assistant can respond on Telegram, Slack, and Discord simultaneously, using the same SOUL.md configuration and accessing the same data sources. One assistant, every platform.

Real Examples

Abstract descriptions only go so far. Here are two concrete assistants that people run in production with OpenClaw.

BI analyst that checks your funnel daily

This assistant connects to Mixpanel and Stripe. Every morning at 9 AM, it pulls the previous day's funnel data: signups, first query, paywall views, trial starts, and paid conversions. It compares each step to the 7-day average and flags anything that dropped more than 15%. The report lands in your Telegram chat before your first coffee.

# Funnel Analyst

## Identity
- Name: Radar
- Role: BI Analyst
- Model: claude-sonnet-4-20250514

## Personality
- Numbers-first, no fluff
- Flags anomalies before being asked
- Uses tables for data presentation

## Rules
- Pull funnel data from Mixpanel every morning at 09:00
- Compare each step to the 7-day rolling average
- Flag any step with a drop greater than 15%
- Include Stripe revenue and active subscription count
- Format the report as a Telegram message with clear sections

## Skills
- mixpanel: Query funnel events and user counts
- stripe: Pull revenue and subscription data
- scheduler: Run tasks on a cron schedule

## Channels
- telegram: true

DevOps agent that monitors uptime

This assistant monitors your web services and alerts you when something goes down. It pings your endpoints every 5 minutes, checks response times, and sends an immediate Slack alert if any endpoint returns an error or takes longer than 3 seconds to respond. It also generates a weekly uptime report with response time trends.

# Uptime Monitor

## Identity
- Name: Sentinel
- Role: DevOps Monitor
- Model: gpt-4o

## Personality
- Concise and urgent when alerting
- Detailed and structured in weekly reports
- Never sends false alarms

## Rules
- Check all endpoints every 5 minutes
- Alert immediately on HTTP 5xx errors
- Alert if response time exceeds 3000ms
- Include response time, status code, and timestamp in alerts
- Generate weekly uptime percentage report every Monday at 08:00
- Endpoints: https://api.example.com/health, https://app.example.com

## Skills
- browser: Ping HTTP endpoints and measure response time
- scheduler: Run health checks on a cron schedule
- file_manager: Store historical response time data

## Channels
- slack: true

Both of these assistants run unattended once configured. You set them up, start the gateway, and they do their job every day without further input. When they find something worth reporting, they come to you.

Running 24/7 on Any Hardware

Your assistant is only useful if it is running. Here are three ways to keep it available around the clock, depending on your budget and technical preference.

Mac (launchd)

If your Mac is always on, you can use launchd to run the OpenClaw gateway as a persistent background service. Create a plist file that starts the gateway at boot and restarts it if it crashes. This is ideal for development and personal use when your machine is plugged in at a desk.

<!-- ~/Library/LaunchAgents/ai.openclaw.gateway.plist -->
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
  "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>ai.openclaw.gateway</string>
  <key>ProgramArguments</key>
  <array>
    <string>/usr/local/bin/openclaw</string>
    <string>gateway</string>
    <string>start</string>
  </array>
  <key>RunAtLoad</key>
  <true/>
  <key>KeepAlive</key>
  <true/>
</dict>
</plist>

VPS (systemd)

A VPS is the most reliable option for 24/7 uptime. A $5/month instance from any cloud provider is enough to run the OpenClaw gateway with multiple agents. Use systemd to manage the process, and your assistant stays online regardless of your personal machine's state.

# /etc/systemd/system/openclaw-gateway.service
[Unit]
Description=OpenClaw Gateway
After=network.target

[Service]
Type=simple
User=openclaw
WorkingDirectory=/home/openclaw/agents
ExecStart=/usr/local/bin/openclaw gateway start
Restart=always
RestartSec=10

[Install]
WantedBy=multi-user.target

Raspberry Pi

A Raspberry Pi is a low-cost, low-power option for running your assistant at home. A Pi 4 or Pi 5 handles the OpenClaw gateway with ease, draws minimal power, and runs silently on your desk or shelf. Install the OpenClaw CLI, set up your agents, and use systemd to keep the gateway running. Total hardware cost is around $75, with no monthly fees beyond your internet connection and LLM API usage.

# Install OpenClaw on Raspberry Pi OS
curl -fsSL https://get.openclaw.com | bash

# Copy your agent configs
scp -r ./agents/ pi@raspberrypi:~/agents/

# Register agents and start
openclaw agents add atlas --workspace ~/agents/atlas
openclaw gateway start

# Enable systemd service for auto-start
sudo systemctl enable openclaw-gateway
sudo systemctl start openclaw-gateway

All three approaches use the same SOUL.md files and the same OpenClaw gateway. Your assistant configuration is portable. You can develop on your Mac, test locally, and deploy to a VPS or Raspberry Pi when you are ready for always-on operation. Read our detailed Raspberry Pi deployment guide and VPS deployment guide for step-by-step instructions.

Build Yours with CrewClaw

Writing a SOUL.md from scratch works, but it takes time to get the structure right, choose the correct skills, and configure the channels properly. CrewClaw generates a complete, production-ready SOUL.md for you in under a minute.

Pick a role from the library or describe your own. Select the integrations your assistant needs. CrewClaw generates the full SOUL.md with identity, personality, rules, skills, and channel configuration. Download the package, run two commands, and your assistant is live.

No trial and error with SOUL.md syntax. No guessing which skills to include. No missing configuration fields. Just a working assistant, ready to deploy.

Frequently Asked Questions