Skip to content

reverseCode23932/discord-ai-bot

BranchesTags

Repository files navigation

Discord AI Bot

Release

A Discord bot with AI chat (OpenAI, Ollama, Groq, or custom), per-user language settings, voice listen (Google + Whisper STT), and Edge TTS in voice channels.

Uses an official Discord bot token from the Developer Portal — not a personal account token.

Features

  • Slash commands: /ask, /say, /askvoice, /listen, /stoplisten, /language, /synthesizer, /voice, /settings, /reset, /leave
  • Voice listen (/listen): speech → hybrid STT → Ollama/LLM → TTS in voice (no chat spam by default)
  • Prefix commands: !ask, !lang, !reset
  • Mention or prefix triggers in text channels
  • Per-user language (en, ru, uk, de, fr, es, pt, ja, zh, pl, tr)
  • Two TTS engines: edge (neural) and gtts (Google)
  • Per-channel conversation memory
  • Rotating logs in data/logs/bot.log

Requirements

  • Python 3.10+
  • FFmpeg (for voice playback) — install guide
  • Discord bot application + token
  • LLM backend: Ollama (free/local), Groq (free tier), or OpenAI

Quick start

git clone https://github.com/reverseCode23932/discord-ai-bot.git
cd discord-ai-bot

python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
# Voice also needs FFmpeg on PATH and davey (included in requirements.txt)

copy .env.example .env
# Edit .env — add DISCORD_BOT_TOKEN and LLM settings (see below)

python run.py

Discord setup

1. Create the application

  1. Open Discord Developer Portal → New Application
  2. Bot → Reset Token → copy into .env as DISCORD_BOT_TOKEN
  3. Under Privileged Gateway Intents, enable:
    • Message Content Intent (required for @mentions and ! prefix commands)

2. Bot permissions (server)

Enable these Bot Permissions when generating the invite link:

Permission Why
View Channels See channels to reply
Send Messages Post AI replies
Read Message History Context in channels
Connect Join voice for TTS
Speak Play TTS audio in voice

Optional (not required): Embed Links, Attach Files.

Permission integer (for manual invite URLs): 3214336

3. OAuth2 scopes

In OAuth2 → URL Generator, select:

Scope Required
bot Yes
applications.commands Yes (slash commands)

Then select the five permissions above and open the generated URL to invite the bot to your server.

Example invite URL shape:

https://discord.com/api/oauth2/authorize?client_id=YOUR_CLIENT_ID&permissions=3214336&scope=bot%20applications.commands

Replace YOUR_CLIENT_ID with Application ID (General Information).

4. Make it feel personal (optional)

In the Developer Portal → Bot, set username and avatar to match your style. This is the supported way to customize appearance — do not use a user account token.

Environment variables

Copy .env.example to .env:

Variable Required Description
DISCORD_BOT_TOKEN Yes Bot token from Developer Portal
LLM_PROVIDER No ollama (default), groq, openai, custom
LLM_MODEL No e.g. llama3.2, llama-3.1-8b-instant, gpt-4o-mini
LLM_API_KEY If groq/openai API key (Groq or OpenAI)
LLM_BASE_URL If custom/ollama e.g. http://127.0.0.1:11434/v1
BOT_PREFIX No Default !
DEFAULT_LANGUAGE No Default en for new users
DEFAULT_SYNTHESIZER No edge or gtts
LOG_LEVEL No INFO or DEBUG
LOG_TO_FILE No true / false
LOG_MAX_BYTES No Max log file size (bytes)
VOICE_WAKE_WORDS No Comma-separated wake phrase(s) for /listen
VOICE_REPLY_TTS No Speak AI replies in voice (true/false)
VOICE_REPLY_TEXT No Post heard text + reply in chat (false default; voice-only)
CHAT_WHISPER_DELETE_AFTER No Auto-delete voice chat replies after N seconds (90)
CHAT_EPHEMERAL No Slash replies only visible to you (true)
STT_ENGINE No local (default), google, openai, auto
WHISPER_LOCAL_MODEL No small (default), base, medium, tiny
WHISPER_DEVICE No cpu or cuda (GPU)
WHISPER_COMPUTE_TYPE No int8 on CPU, float16 on GPU
WHISPER_BEAM_SIZE No Higher = more accurate, slower (default 5)
WHISPER_VAD_FILTER No Skip silence/noise (true)
LLM_TIMEOUT_SECONDS No Max wait for Ollama/Groq reply (default 180)
MIN_SPEECH_BYTES No Min audio before STT (default ~0.4s)

Commands

Command Description
/ask Ask the AI (your language)
/say Speak text in your voice channel
/askvoice AI reply + speak in voice
/language Set or list languages
/synthesizer edge or gtts
/voice Pick Edge TTS voice
/settings Show your preferences
/reset Clear channel chat history
/listen Join your VC and listen for spoken commands
/stoplisten Stop voice listening
/leave Disconnect from voice and stop listening

LLM without OpenAI (quota / billing)

Option A — Ollama (local, free)

  1. Install Ollama

  2. In a terminal:

    ollama serve
ollama pull llama3.2

  3. In .env:

    LLM_PROVIDER=ollama
LLM_MODEL=llama3.2
STT_ENGINE=local

Option B — Groq (cloud, free tier)

  1. Get a key at https://console.groq.com

  2. In .env:

    LLM_PROVIDER=groq
LLM_API_KEY=gsk_your_key_here
LLM_MODEL=llama-3.1-8b-instant

Option C — LM Studio (local GUI)

  1. Load a model in LM Studio → start server

  2. In .env:

    LLM_PROVIDER=custom
LLM_BASE_URL=http://127.0.0.1:1234/v1
LLM_API_KEY=lm-studio
LLM_MODEL=name-shown-in-lm-studio

Voice commands (/listen)

  1. Join a voice channel.
  2. Run /listen in a text channel (same server).
  3. Wait for the bot to join, then speak when Discord shows your green speaking indicator.
  4. The bot transcribes speech → asks the LLM → speaks the reply in voice (no chat spam by default).

Optional in .env:

  • VOICE_WAKE_WORDS=hey bot — only react after that phrase (comma-separated for multiple).
  • VOICE_WAKE_WORDS= (empty) — react to any speech.
  • Say "стоп" / "stop listening" or use /stoplisten to end.

Requires: discord-ext-voice-recv, davey, FFmpeg, and faster-whisper (local STT — no OpenAI key).

Speech recognition tips (Whisper)

Voice recognition uses Google Speech + Whisper (STT_ENGINE=hybrid by default) — not Ollama. Ollama only generates text replies after speech is already transcribed. Discord audio is converted to mono 16 kHz with volume boost before STT.

Pick a model size

Model Speed Russian accuracy RAM (approx.)
tiny Fastest Poor (hallucinations on silence) ~1 GB
base Fast OK ~1 GB
small Medium Good balance ~2 GB
medium Slower Clearest (default, RTX GPU) ~5 GB

In .env:

STT_ENGINE=local
WHISPER_LOCAL_MODEL=small
WHISPER_VAD_FILTER=true
WHISPER_BEAM_SIZE=5

First run downloads the model (~500 MB for small). Check logs:

Loading local Whisper 'small' (device=cpu compute=int8...)
local STT (12 chars): привет как дела

.env gotchas

  • Do not duplicate WHISPER_LOCAL_MODEL — if it appears twice, only the first line wins (dotenv default). Keep one line.

  • Set /language ru (or your code) so Whisper uses the right language hint.

  • If recognition is still weak, try WHISPER_LOCAL_MODEL=medium (slower, ~5 GB RAM).

  • GPU (NVIDIA RTX, e.g. 4060): install CUDA DLLs into your venv, then use cuda:

    .\.venv\Scripts\activate   # or: C:\Users\Gleb\Envs\dsbot\Scripts\activate
.\scripts\install-gpu-stt.ps1
    WHISPER_DEVICE=cuda
WHISPER_COMPUTE_TYPE=float16
WHISPER_LOCAL_MODEL=small

    On startup you should see: NVIDIA CUDA DLL paths added and device=cuda compute=float16. If CUDA still fails, the bot falls back to CPU automatically.

Speaking tips

  • Use a headset mic; avoid music or TV in the background.
  • Speak 1–2 seconds after the green ring appears; pause briefly between phrases.
  • Wait for Ollama on CPU — the first reply can take 30–60+ seconds; do not /stoplisten too early.

Alternative: Google STT (no local model download)

Requires internet; often good for short Russian phrases:

STT_ENGINE=google

Install: pip install SpeechRecognition

Project structure

discord-ai-bot/
├── run.py                 # Entry point
├── requirements.txt
├── .env.example
├── data/                  # Runtime data (gitignored)
│   ├── user_settings.json
│   └── logs/bot.log
└── src/discord_ai/
    ├── main.py
    ├── config.py
    ├── logging_setup.py
    ├── bot/client.py
    ├── commands/          # slash + prefix
    ├── services/          # ai, voice, listen, stt, settings, history
    ├── i18n/languages.py
    └── tts/synthesizer.py

Logging

Logs go to the console and data/logs/bot.log (rotating, 5 backups).

Set LOG_LEVEL=DEBUG to log full prompts and replies.

Troubleshooting

Issue Fix
Bot does not reply to !ask Enable Message Content Intent in Developer Portal
Slash commands missing Re-invite with applications.commands scope; wait ~1 min after start
Voice error 4017 / cannot join VC pip install davey and discord.py>=2.7
/listen does nothing Install discord-ext-voice-recv; speak when green ring shows
OpusError: corrupted stream Reinstall deps from requirements.txt (DAVE-patched voice-recv fork)
Voice silent / no audio Install FFmpeg and ensure it is on PATH; wait for LLM (check logs for LLM reply)
Wrong / nonsense transcription Use WHISPER_LOCAL_MODEL=small or medium; remove duplicate .env lines; see Speech recognition tips
Hallucinations like «Редактор субтитров» Normal for tiny on silence — upgrade model; bot filters common phrases
Missing env vars Create .env from .env.example
OpenAI 429 / quota Switch to LLM_PROVIDER=ollama or groq; use STT_ENGINE=local
Ollama connection / no voice reply Run ollama serve and ollama pull llama3.2; increase LLM_TIMEOUT_SECONDS
cublas64_12.dll is not found Run .\scripts\install-gpu-stt.ps1 in your venv, then WHISPER_DEVICE=cuda
Log spam unexpected rtcp packet Harmless; suppressed at default log levels in recent versions
gh not recognized Close/reopen terminal or use full path to gh.exe (see below)

GitHub — clone, update, and push

Clone this repo

git clone https://github.com/reverseCode23932/discord-ai-bot.git
cd discord-ai-bot
python -m venv .venv
.\.venv\Scripts\activate
pip install -r requirements.txt
copy .env.example .env
# Edit .env — never commit real tokens

Daily workflow (after you change code)

cd C:\path\to\discord-ai-bot
git status
git add .
git commit -m "describe your change"
git push

Never commit .env — it is gitignored. Only commit .env.example with placeholders.

First-time publish (new fork)

  1. Install GitHub CLI (winget install GitHub.cli)

  2. Log in:

    cd C:\Users\Gleb\discord-ai-bot
.\scripts\gh-login.ps1

    Or:

    & "C:\Program Files\GitHub CLI\gh.exe" auth login

    If gh is not recognized: close and reopen PowerShell, or use the scripts in scripts/.

  3. Create repo and push:

    .\scripts\publish-to-github.ps1

    Or manually:

    git remote add origin https://github.com/<your-username>/discord-ai-bot.git
git branch -M main
git push -u origin main

Pull updates from GitHub

git pull origin main
pip install -r requirements.txt

If you use a venv in another folder (e.g. Envs\dsbot), activate it and reinstall there too.

Issues & contributions on GitHub

  • Bug reports: open an Issue with logs from data/logs/bot.log (remove your bot token first).
  • Pull requests: fork → branch → commit → PR. Keep changes focused; do not include .env or data/.

Releases

See Releases on GitHub.
Latest: v1.0.0 — first stable build.

git clone https://github.com/reverseCode23932/discord-ai-bot.git
cd discord-ai-bot
git checkout v1.0.0

License

MIT — see LICENSE.

About

No description, website, or topics provided.

Resources

Readme

License

MIT license
Activity

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases 1

v1.0.0 Latest
May 16, 2026

Packages

 
 
 

Contributors

Languages