Deploy a Multi-Channel Agent

In this lab, you will configure and deploy a multi-channel AI agent that can respond to users across Telegram and Discord simultaneously. You will set up the agent configuration, build bot handlers for each platform, create a systemd service for 24/7 operation, and implement a monitoring script to keep everything running smoothly.

This is a real-world deployment pattern used by production AI agents. By the end of this lab, you will have a complete deployment blueprint that you can adapt to your own projects.

Architecture Overview

+-------------------+     +-------------------+
|   Telegram Bot    |     |   Discord Bot     |
|  (telegram-bot.py)|     |  (discord-bot.py) |
+--------+----------+     +--------+----------+
         |                          |
         v                          v
+-------------------------------------------+
|          Agent Core (config.yaml)          |
|    Model selection, API keys, settings     |
+-------------------------------------------+
         |
         v
+-------------------------------------------+
|       systemd Service (systemd-service.conf)|
|    Auto-restart, logging, process mgmt     |
+-------------------------------------------+
         |
         v
+-------------------------------------------+
|      Monitoring (monitoring.py)            |
|    Health checks, log parsing, alerts      |
+-------------------------------------------+

Step 1: Agent Configuration (config.yaml)

Create a central configuration file that both bots will read from:

Model Configuration:

• Select a primary LLM model for generating responses
• Select a fallback model in case the primary is unavailable
• Set temperature, max tokens, and other generation parameters

API Keys (Placeholders):

• Telegram bot token placeholder
• Discord bot token placeholder
• LLM API key placeholder
• Use environment variable references (e.g., ${TELEGRAM_BOT_TOKEN}) rather than hardcoded values

Deployment Settings:

• Deployment mode (development, staging, production)
• Log level (debug, info, warning, error)
• Rate limiting settings (max requests per user per minute)
• Response timeout in seconds

Channel Configuration:

• For each channel (Telegram, Discord): enabled/disabled flag, allowed chat IDs or channel IDs, command prefix

Step 2: Telegram Bot Handler (telegram-bot.py)

Build a Telegram bot using the python-telegram-bot library pattern:

Required Handlers:

• /start command: Send a welcome message introducing the agent and its capabilities
• /help command: List all available commands and usage instructions
• Message handler: Process incoming text messages, send them to the LLM, and return the response
• Error handler: Catch and log errors gracefully without crashing the bot

Implementation Details:

• Load configuration from config.yaml using PyYAML
• Implement rate limiting per user (check config for max requests per minute)
• Add logging with timestamps for every incoming message and outgoing response
• Include a process_with_llm(message: str) -> str function that sends the user message to the configured LLM
• Handle long responses by splitting them into multiple messages (Telegram has a 4096-character limit)

Bot Setup:

• Use ApplicationBuilder from python-telegram-bot to create the bot application
• Register all handlers with appropriate filters
• Include a main() function that starts polling

Step 3: Discord Bot Handler (discord-bot.py)

Build a Discord bot using the discord.py library pattern:

Required Events:

• on_ready: Log that the bot is connected and list the servers it has joined
• on_message: Process incoming messages, ignore bot's own messages, send to LLM, and reply
• Channel routing: Only respond in configured channels (check config for allowed channel IDs)

Implementation Details:

• Load configuration from config.yaml using PyYAML
• Implement rate limiting per user (check config for max requests per minute)
• Add logging with timestamps for every incoming message and outgoing response
• Include a process_with_llm(message: str) -> str function (same pattern as Telegram bot)
• Handle long responses by splitting into multiple messages (Discord has a 2000-character limit)
• Use commands.Bot with a configurable command prefix from config

Commands:

• !help (or configured prefix + help): List available commands
• !status: Show bot uptime and number of messages processed
• Regular messages in allowed channels: Process with LLM and reply

Step 4: systemd Service File (systemd-service.conf)

Create a systemd service unit file for running the agent 24/7 on a Linux server:

Service Configuration:

• [Unit] section: Description, documentation URL, dependency on network
• [Service] section:
  • Working directory for the agent code
  • Python executable path and entry point script
  • Environment file for loading API keys (e.g., /etc/agent/.env)
  • Restart policy: on-failure with a 10-second delay
  • Maximum restart attempts (5 restarts within 300 seconds)
  • Logging to stdout/stderr (captured by journald)
  • Resource limits: memory limit, CPU quota
  • Security hardening: NoNewPrivileges=true, ProtectSystem=strict
• [Install] section: Enable on multi-user target

Step 5: Monitoring Script (monitoring.py)

Build a health check and monitoring script:

Health Checks:

• Check if the Telegram bot process is running (by PID file or process name)
• Check if the Discord bot process is running
• Check if the LLM API is reachable (HTTP health check with timeout)
• Report overall system health status

Log Parsing:

• Read the last N lines from the agent log file
• Count errors and warnings in the recent logs
• Extract the timestamp of the last successful response
• Calculate error rate (errors per hour)

Alerting:

• Define alert thresholds (e.g., error rate > 10/hour, process down, API unreachable)
• Print alert messages to stdout with severity levels
• Include a send_alert(message: str, severity: str) function that can be extended to send notifications (email, Slack, etc.)

Report Generation:

• Generate a health report with: uptime, messages processed, error count, API latency, memory usage
• Include a main() function that runs all checks and prints the report

What to Submit

The editor has 5 file sections with TODO comments. Replace each TODO with your code. The AI grader will evaluate each section against the rubric.

Hints

• For config.yaml, use a clean hierarchical structure. Group related settings under top-level keys like model, telegram, discord, deployment
• For the Telegram bot, the python-telegram-bot library uses Application.builder().token(TOKEN).build() pattern for v20+
• For the Discord bot, discord.py uses commands.Bot(command_prefix="!", intents=intents) pattern
• For systemd, remember that environment variables from .env files use EnvironmentFile= directive
• For monitoring, psutil is a common library for process checks, but you can also use subprocess with pgrep