Channel Configuration¶

This guide shows practical Telegram and Discord setups with both interactive and headless CLI commands.

Interactive setup (default)¶

For human-operated setup, run these commands and answer prompts:

llm-expose add model
llm-expose add channel
llm-expose add pair
llm-expose start

Use --no-input only for automation. In headless mode, missing required values fail immediately.

Telegram setup¶

Step 1: Create a bot token¶

Open Telegram and start a chat with @BotFather.
Run /newbot and follow the prompts.
Copy the bot token (123456789:AA...).

Step 2: Add or reuse a model¶

llm-expose add model \
  --name gpt4o-mini \
  --provider openai \
  --model-id gpt-4o-mini \
  -y --no-input

Step 3: Add the Telegram channel config¶

llm-expose add channel \
  --name support-telegram \
  --client-type telegram \
  --bot-token "123456789:AAExampleTelegramToken" \
  --model-name gpt4o-mini \
  --system-prompt-path ./prompts/support.txt \
  -y --no-input

Step 4: Pair allowed chat IDs¶

llm-expose add pair 123456789 --channel support-telegram --no-input

Use one add pair per allowed user/group chat ID.

Discord setup¶

Step 1: Create a bot token¶

Open Discord Developer Portal.
Create or select an application, then create a bot.
Enable Message Content Intent in Bot -> Privileged Gateway Intents.
Copy the bot token.

Step 2: Add the Discord channel config¶

llm-expose add channel \
  --name ops-discord \
  --client-type discord \
  --bot-token "YOUR_DISCORD_BOT_TOKEN" \
  --model-name gpt4o-mini \
  --system-prompt-path ./prompts/ops.txt \
  -y --no-input

Step 3: Pair allowed channel IDs¶

llm-expose add pair 987654321098765432 --channel ops-discord --no-input

Use one add pair per Discord channel ID that should receive replies.

Attach MCP servers to a channel¶

Attach MCP servers at channel creation time by repeating --mcp-server:

llm-expose add channel \
  --name ops-discord \
  --client-type discord \
  --bot-token "YOUR_DISCORD_BOT_TOKEN" \
  --model-name gpt4o-mini \
  --mcp-server web-search \
  -y --no-input

If an MCP server name does not exist, headless mode fails with an explicit error.

Example channel YAMLs¶

Telegram channel config:

client_type: telegram
bot_token: "123456789:AAExampleTelegramToken"
mcp_servers:
  - web-search
system_prompt_path: ./prompts/support.txt
model_name: gpt4o-mini

Discord channel config:

client_type: discord
bot_token: "YOUR_DISCORD_BOT_TOKEN"
mcp_servers:
  - web-search
system_prompt_path: ./prompts/ops.txt
model_name: gpt4o-mini

Prompt file strategy¶

Keep prompt files in a dedicated directory like ./prompts/.
Use one prompt per channel role (support.txt, ops.txt, oncall.txt).
If system_prompt_path is missing or unreadable at startup, runtime falls back to the default system prompt.

Start and verify¶

llm-expose start --channel support-telegram -y --no-input
# or
llm-expose start --channel ops-discord -y --no-input

Verify persisted configuration:

llm-expose list channels
llm-expose list pairs
llm-expose list models

Troubleshooting¶

Telegram reports another polling process: stop any other process using the same bot token.
Discord bot receives empty content: confirm Message Content Intent is enabled.
Runtime says channel is not paired: add pair IDs for the exact chat/channel where messages arrive.
Model is not used: confirm model_name points to an existing model config.