🎧🤖 The Local-First Discord Bot

by | Feb 1, 2026 | Ai, coding, music, tutorials | 0 comments

Goal: build a Discord bot that does the normal bot things people actually use — 🎧 music controls, 🧰 slash commands, and (optionally) 🤖 a local AI “/ask” command powered by LM Studio.

This isn’t “AI sci-fi”. This is a practical automation project: your PC does the boring parts, and Discord gets a bot that feels responsive and useful.

Rule I use: ⏱️ If the bot can’t respond in 10 seconds (or show it’s working), it doesn’t work.

✅ What this bot can do (v1)

  • Music: /join, /play, /pause, /resume, /skip, /queue, /leave 🎵
  • Utility: /ping, /help, /roll 🧰
  • Optional: /ask → routes to LM Studio running locally 🤖

The vibe: a bot that feels “premium” because it’s reliable — no mystery silence, no broken replies, no random crashes after one song.

🧠 The simple architecture

Discord (slash command)
  └─> Your bot (Node)
       ├─> Voice engine (@discordjs/voice + FFmpeg)  🎧
       └─> Optional AI engine (LM Studio /v1)        🤖

If you’re new: don’t panic. You’re basically building a small server app that listens for Discord commands and runs two jobs:
play audio, and (optionally) call a local text model.

🔗 The 3 links you actually need

🎛 Step 1 — Create the Discord bot

  1. Go to the Discord Developer Portal → create an application.
  2. Add a bot user.
  3. Copy the bot token (keep it private).
  4. Invite it to your server with scopes: bot + applications.commands.
  5. Give it permissions: Send Messages, Use Slash Commands, Connect, Speak.

Tip: Slash commands are cleaner than “message commands” because you don’t need to enable message content intents just to get started.

🤖 Step 2 (optional) — Turn your PC into the bot’s “brain” with LM Studio

LM Studio can run a model locally and expose an OpenAI-compatible endpoint. That means your bot can send a normal chat completion request to:

http://127.0.0.1:1234/v1/chat/completions

Why this matters: you’re not waiting on a cloud service, you’re not juggling accounts, and you control the model + the machine.

Receipt: when this works, you can type /ask in Discord and the reply comes from your own PC.

🎧 Step 3 — Music playback without pain

Here’s the beginner truth: the “music bot” part is mostly about managing a queue and handling the player lifecycle properly.

✅ The core pattern (what makes it reliable)

  • One queue per server (guild)
  • One audio player per server
  • A playNext() function that always knows what to do next
  • When the player goes Idle, call playNext() again

If you miss the Idle handler, you get the classic failure: “it played one thing and stopped.”

Rule: If your bot can’t play the next track automatically, it’s not a music bot — it’s a demo.

⚠️ The music source gotcha (start stable)

If you want your first version to actually work, use sources that don’t constantly break:

  • ✅ direct MP3/OGG stream URLs (internet radio)
  • ✅ your own hosted audio files
  • ✅ local files you control

Save the “hard mode” sources for later. The goal is a bot that works every day, not a bot that works once.

🪲 Gotchas and fixes (the ones that waste hours)

1) “Interaction failed”

Cause: Discord expects a response quickly. If your command takes time, you must acknowledge it first.

Fix: defer the reply immediately, then edit the reply when you finish. (This is standard slash command behaviour.)

2) Bot joins voice but plays silence 🔇

Cause: missing Opus or FFmpeg, or the audio resource isn’t built correctly.

Fix: follow the discord.js voice setup and verify dependencies using their voice guide.

3) Plays first track then stops

Cause: no “Idle → playNext()” loop.

Fix: wire the player status events so the queue keeps moving.

4) /play given a webpage URL

Cause: a webpage is not an audio stream.

Fix: start with direct streams, then upgrade later.

🧾 The “Receipts” checklist (so you know it’s real)

Record a quick screen capture and prove it:

  1. /join → bot joins voice 🎧
  2. /play a direct audio stream → audio plays ✅
  3. /queue → shows queue state 🧾
  4. (optional) /ask → bot replies from LM Studio 🤖

That’s the moment where it stops being “AI hype” and starts being your machine doing useful work.

🚀 Next upgrades (if you want it to feel premium)

  • 🔐 Role-based command locking (mods only for skip/clear queue)
  • 🧠 Short memory for /ask (last 6–10 messages)
  • 📌 “Now playing” embed + progress
  • 🧰 Tool-style AI: the model can request “show queue”, “now playing”, etc. (LM Studio supports tool calling)

If you want, I’ll publish a “v2” post that includes the full ready-to-run project structure, command registration, and the exact code you can copy into a folder and run.