Axiomkit

Overview

This example shows how to connect an Axiom agent to Discord using @axiomkit/discord. You will create a simple bot that replies in channels or DMs using a Groq model.

Prerequisites

Discord Bot Token: Create a bot in the Discord Developer Portal, add a bot user, and copy the token (DISCORD_BOT_TOKEN).
Bot Name: Set the name you want your bot to use (DISCORD_BOT_NAME).
Groq API Key: Create a Groq account and generate an API key (GROQ_API_KEY).

Install

Using pnpm:

pnpm add @axiomkit/core @axiomkit/discord @axiomkit/cli @ai-sdk/groq zod

Environment

Set the following environment variables in file .env before running your bot:

DISCORD_BOT_TOKEN=YOUR_DISCORD_BOT_TOKEN
DISCORD_BOT_NAME=YOUR_BOT_DISPLAY_NAME
GROQ_API_KEY=YOUR_GROQ_API_KEY
# Optional: enable attachment processing in inputs/outputs
PROCESS_ATTACHMENTS=false

Minimal bot

Create a file like example-discord.ts with the following content:

import {
  createContainer,
  createAgent,
  LogLevel,
  validateEnv,
  Logger,
  context,
} from "@axiomkit/core";
import * as z from "zod/v4";
import { discord } from "@axiomkit/discord";
import { groq } from "@ai-sdk/groq";

// Validate environment before proceeding
const env = validateEnv(
  z.object({
    GROQ_API_KEY: z.string().min(1, "GROQ_API_KEY is required"),
    DISCORD_BOT_TOKEN: z.string().min(1, "DISCORD_BOT_TOKEN is required"),
    DISCORD_BOT_NAME: z.string().min(1, "DISCORD_BOT_NAME is required"),
  })
);

const container = createContainer();
const discordContext = context({
  type: "discord-context",
  instructions: `You are a helpful Discord bot assistant. Your Name is ${env.DISCORD_BOT_NAME}. You will respond to messages in a friendly and helpful manner.

IMPORTANT: When responding to Discord messages, always output clean, natural text. Do NOT output JSON or code blocks unless specifically requested.

Guidelines:
- Keep responses conversational and friendly
- Use Discord formatting sparingly (bold, italic, etc.) when appropriate
- If someone asks for help with code, provide clear explanations
- For long responses, break them into multiple messages naturally
- Use emojis occasionally to make responses more engaging
- If someone shares an image or file, acknowledge it appropriately
- Be helpful but concise
- Don't Reply JSON like  {"content": "Hey !  What's up?"}
  
Current context: You're in a Discord channel. Respond naturally as if chatting with friends.`,
});
const agent = createAgent({
  logger: new Logger({ level: LogLevel.DEBUG }),
  context: discordContext,
  model: groq("gemma2-9b-it"),
  providers: [discord],
  container,
});

console.log("Starting Axiomkit Discord Bot...");
await agent.start({
  id: "discord-bot",
});
console.log("Axiomkit Discord Bot started");

Run the script with your preferred TypeScript runner (e.g., tsx, ts-node, or bun):

npx tsx example-discord.ts

How messages flow

Input: Incoming Discord messages are mapped to input discord:message with params userId, username, and a flag hasAttachments. The input data contains text and optional attachments.
Output (text): Your agent can send messages by emitting output discord:message with a plain string.
Output (attachments): If PROCESS_ATTACHMENTS=true, your agent can emit discord:message-with-attachments with a JSON payload containing content and an attachments array of { url, filename?, description? }.

Example outputs your agent can produce:

<output type="discord:message">Hi!</output>

<output type="discord:message-with-attachments">
  {
    "content": "Here's the image you requested!",
    "attachments": [
      {
        "url": "https://example.com/image.jpg",
        "filename": "result.jpg",
        "description": "Generated image"
      }
    ]
  }
</output>

Notes

Messages longer than ~1500 characters are automatically split and sent in multiple parts.
DMs and guild text channels are supported.

Discord