Phase 1 / Ep 06: Your First Message — The Complete Conversation Flow from CLI to Telegram

🎯 Learning Objective: Successfully send your first message and understand the complete message flow path.

1. Direct Conversation via CLI

The simplest way—talk to the Agent directly in the terminal:

# Start interactive chat
openclaw chat

# Or single message mode
openclaw chat "What's the weather like today?"

In interactive mode, you will see:

🦀 OpenClaw Chat (personal)
Type /help for commands, /quit to exit
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

You: Hello, introduce yourself

Agent: Hello! I am your OpenClaw personal Agent.
I run on your local machine and can help you:
- Search the internet
- Manage files
- Execute Shell commands
- Set reminders
How can I help you today?

2. Telegram Integration

Step 1: Create a Bot with BotFather

1. Search for @BotFather in Telegram
2. Send /newbot
3. Follow the prompts to fill in the Bot name and username
4. Obtain the Bot Token: 7123456789:AAH...

Step 2: Add Telegram Channel

openclaw channels add telegram --token "7123456789:AAH..."

Step 3: Pairing

Send any message to your Bot, and the Agent will return a Pairing Code. Confirm it in the CLI:

openclaw channels pair <code>

Pairing complete! Now you can talk to the Agent directly in Telegram.

3. End-to-End Message Flow

sequenceDiagram
    actor User as 🧑 User
    participant TG as 📱 Telegram Bot
    participant GW as 🦀 Gateway
    participant Auth as 🔐 Auth Engine
    participant Perm as 🛡️ Permission Check
    participant Agent as 🤖 Agent
    participant Boot as 📋 Bootstrap
    participant Mem as 🧠 Memory
    participant LLM as 🧪 Claude API

    User->>TG: "Help me check tomorrow's weather in Beijing"
    TG->>GW: Webhook Push
    GW->>Auth: Verify user identity
    Auth->>Perm: Check permission level
    Perm->>Agent: Route to personal Agent

    Note over Agent: Initialize context
    Agent->>Boot: Load SOUL.md, USER.md...
    Agent->>Mem: Retrieve relevant memory
    Agent->>LLM: Assemble Prompt + Tool definitions

    LLM->>Agent: Return tool_use: web_search
    Agent->>Agent: Execute web_search("Beijing tomorrow weather")
    Agent->>LLM: Submit tool execution results
    LLM->>Agent: Return final response

    Agent->>GW: Pass response
    GW->>TG: Push reply
    TG->>User: "Tomorrow in Beijing: Sunny, 28°C / 17°C\nAir Quality: Good"

4. CLI vs Telegram Comparison

5. Congratulations!

Your Agent is now alive. It runs on your machine, possesses the identity you gave it (SOUL.md), knows who you are (USER.md), and is learning to remember your conversations.

Starting from the next chapter, we will dive deep into the CLI command system—mastering every OpenClaw command to become a true Agent administrator.

Next Episode Preview: Ep 07, entering Phase 2! We start with Gateway lifecycle management: start, stop, restart, and daemon mode. Each command will synchronously demonstrate its equivalent operation in both CLI and Telegram.