Send Your First Message - First Conversation with AI Assistant

What You'll Learn

After completing this tutorial, you'll be able to:

• Send messages to your AI assistant via CLI
• Understand OpenClaw's message sending mechanism and session management
• Use various parameters to customize message sending behavior
• Have conversations with your Agent across different channels

Your Current Situation

The Gateway service is running, but you might be wondering:

• How do I actually have a conversation with the AI assistant?
• What are the ways to send messages?
• How do I manage different conversation sessions?

Core Concepts

OpenClaw provides multiple ways to interact with your AI assistant:

┌─────────────────────────────────────────────────────────────┐
│                    OpenClaw Message Interaction             │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│   CLI Method                       Channel Method           │
│   ┌──────────────┐                 ┌──────────────┐        │
│   │ openclaw     │                 │ WhatsApp     │        │
│   │   agent      │                 │ Telegram     │        │
│   │   message    │                 │ Discord      │        │
│   └──────┬───────┘                 └──────┬───────┘        │
│          │                                │                │
│          └──────────────┬─────────────────┘                │
│                         │                                  │
│                         ▼                                  │
│              ┌────────────────────┐                        │
│              │   Gateway Service  │                        │
│              │   Session Mgmt     │                        │
│              └─────────┬──────────┘                        │
│                        │                                   │
│                        ▼                                   │
│              ┌────────────────────┐                        │
│              │   AI Agent (Pi)    │                        │
│              │   Process & Reply  │                        │
│              └────────────────────┘                        │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Session Management Concepts

In OpenClaw, every conversation belongs to a Session:

• Sessions are identified by a unique sessionKey
• Session states are persisted in the ~/.openclaw/sessions/ directory
• Multiple sessions run in parallel, each maintaining its own context independently

Hands-On Guide

Step 1: Send Messages Using the Agent Command

Why
The openclaw agent command is the primary way to directly talk to your AI assistant.

Open your terminal and type the following commands:

# Basic usage - send a message to the default Agent
openclaw agent --message "Hello, please introduce yourself"

# Or use the shorthand
openclaw agent -m "Hello"

What You Should See
The AI assistant will process your message and return a response, similar to:

┌─────────────────────────────────────┐
│  AI Assistant (claude-3-5-sonnet)   │
├─────────────────────────────────────┤
│  Hello! I'm your AI assistant,      │
│  ready to interact with you through │
│  various messaging channels. I can  │
│  help with coding, file operations, │
│  web browsing, and more.            │
└─────────────────────────────────────┘

Step 2: Specify the Target Session

Why
You may need to talk to a specific session or phone number.

# Specify phone number (E.164 format)
openclaw agent --to "+86138xxxxxxxx" --message "Hello"

# Specify session ID
openclaw agent --session-id "my-session-001" --message "Let's continue our previous topic"

Step 3: Adjust Thinking Depth

Why
Different tasks require different thinking depths. Complex problems may benefit from higher thinking levels.

# Turn off thinking - direct response
openclaw agent --message "Simple question" --thinking off

# Minimal thinking - very light reasoning
openclaw agent --message "Quick question" --thinking minimal

# Low thinking depth - fast response
openclaw agent --message "Simple question" --thinking low

# Medium thinking depth (default)
openclaw agent --message "General question" --thinking medium

# High thinking depth - deep reasoning
openclaw agent --message "Complex problem" --thinking high

Thinking Level Comparison

Level	Use Case	Response Speed
off	Direct answers without reasoning	Fastest
minimal	Very light reasoning	Very fast
low	Simple Q&A, casual chat	Fast
medium	General tasks	Medium
high	Complex reasoning, coding	Slower

Step 4: Use the Message Command

Why
The openclaw message command provides richer message operation capabilities.

# Send a regular message
openclaw message send --message "Hello"

# Send to a specific target
openclaw message send --to "+86138xxxxxxxx" --message "Hello"

# Use JSON output format
openclaw message send --message "Hello" --json

# Dry run (simulate without actually sending)
openclaw message send --message "Test" --dry-run

Step 5: Send Images

Why
AI can understand and analyze image content.

# Send a message with an image
openclaw agent --message "Analyze this image" --images ./photo.png

# Multiple images
openclaw agent --message "Compare these two images" --images ./img1.png --images ./img2.png

Checkpoint

Run the following commands to verify your setup:

# Check if Gateway is running
openclaw status

# You should see output similar to
┌─────────────────────────────────────┐
│  Gateway Status: running            │
│  Port: 18789                        │
│  Auth: token                        │
└─────────────────────────────────────┘

Troubleshooting

Common Errors

1. Gateway Not Running
   Error: Error: Gateway is not running
   Solution: Run openclaw gateway run first

2. Invalid Phone Number Format
   Error: Invalid E.164 format
   Solution: Use +86138xxxxxxxx format, including country code

3. Agent Doesn't Exist
   Error: Unknown agent id "xxx"
   Solution: Run openclaw agents list to see available Agents

4. Thinking Level Not Supported
   Error: Thinking level "xhigh" is only supported for...
   Solution: Check if your current model supports xhigh thinking level

Summary

In this tutorial, you learned:

• ✅ Use the openclaw agent command to send messages
• ✅ Specify target sessions and phone numbers
• ✅ Adjust thinking depth to control response quality
• ✅ Use openclaw message to send messages
• ✅ Pass images to AI for analysis

Next Steps

In the next tutorial, we'll learn Configuration Basics.

You'll learn:

• The structure of openclaw.json config file
• How to use environment variables
• Detailed explanation of key configuration options

---

Appendix: Source Code Reference

Click to expand and view source locations

Last updated: 2026-02-14

Feature	File Path	Line Numbers
Agent Command Implementation	src/commands/agent.ts	64-529
Message Command Implementation	src/commands/message.ts	14-67
Pi Agent Embedded Runner	src/agents/pi-embedded.ts	-
Session Parsing Logic	src/commands/agent/session.ts	-
Thinking Level Handling	src/auto-reply/thinking.ts	-