Discord Bot Notifier

A provider-agnostic Discord chatbot for opt-in AI model alerts and slash-command conversations. Each server operator brings their own Discord bot, provider keys, and model IDs. The repository also includes an Agent Plugins 1.0 skill and the original one-shot notifier script.

Supported model families: GPT (OpenAI), Claude, Gemini, Mistral, Grok, Llama (local OpenAI-compatible endpoint), and a general Local / OpenAI-compatible endpoint. Mistral and Grok use their OpenAI-compatible chat endpoints; Claude and Gemini use their native APIs. A model ID is an exact value from your provider, not a name this project guesses.

Quick start

1. Create a bot in the Discord Developer Portal and invite it to a server with the bot and applications.commands scopes. Grant View Channel and Send Messages in the target channel. No Message Content intent is needed for slash commands.

2. Clone the repository, create a Python 3.10+ virtual environment, and install dependencies:

   python3 -m venv .venv
   . .venv/bin/activate
   pip install -r requirements.txt
   cp .env.example .env
   

3. Privately fill DISCORD_BOT_TOKEN in .env. For fast slash-command registration, set DISCORD_GUILD_ID to your test server ID. Set DISCORD_ALERT_CHANNEL_ID to the channel where alerts should be posted. Never commit .env or send keys through Discord.

4. Run python3 -m bot.app. Keep the process running for slash commands and alerts.

5. In the server, run /addmodel. Select a model family, enter its exact model ID, and optionally its local/compatible base URL. The bot sends an ephemeral template to copy into .env. Fill API keys on the bot host and restart the bot. /addmodel does not receive or store keys.

6. Run /modelstatus, /turnon model:<provider>, /testmodel model:<provider>, and /ask model:<provider> prompt:<your test>. The test calls the AI and returns its answer via Discord. Model status turns green when its required environment configuration is present; last test status shows whether a request succeeded. Green does not by itself prove connectivity.

For a locally hosted model, use an OpenAI-compatible server such as one listening at http://127.0.0.1:11434/v1. The server must be reachable from the machine running the bot. 127.0.0.1 points to that machine, not to the Discord user's device. Local keys are optional; configure MODEL_LLAMA_API_KEY or MODEL_LOCAL_API_KEY if the local server requires authentication.

Commands

Alert events are test_success, test_failure, request_failure, and recovered. They are off by default. Set DISCORD_ALERT_CHANNEL_ID, then enable the events you want with /alerts. Alerts fire when bot commands produce those events; this release does not poll provider uptime in the background. Model switches, alert choices, and last request results are stored per server in SQLite (DISCORD_STATE_DB). Credentials and model IDs stay in environment settings, never in SQLite.

/ask is restricted to Manage Server to limit unwanted provider charges. Bot replies suppress all mentions, including mentions returned by AI. A provider error is reported without echoing the API key or the provider's response body. The bot does not let an AI autonomously call Discord tools; it handles a user's command, calls the chosen provider, and sends the response back.

Environment variable mapping

You can optionally set MODEL_<FAMILY>_BASE_URL to override any default endpoint, and MODEL_LLAMA_API_KEY or MODEL_LOCAL_API_KEY for authenticated local hosts. Use HTTPS for remote providers. One model ID per family is supported in this release. Restart the bot after editing .env; /addmodel prepares the template but does not modify the bot host.

Standalone notifier

The original sender remains available for a one-off Discord alert, even without the chatbot runtime:

python3 skills/send-discord-notification/scripts/send_discord.py --dry-run --message '✅ Job completed'
printf '%s' '✅ Job completed' | python3 skills/send-discord-notification/scripts/send_discord.py --message-stdin

It reads DISCORD_BOT_TOKEN and DISCORD_CHANNEL_ID from the process environment. It does not load .env itself. A successful send returns a Discord message ID; after a network timeout, verify the channel before retrying to avoid duplicates.

Plugin and source layout

• plugin.json and skills/send-discord-notification/ provide the assistant workflow and one-off sender.
• bot/ contains slash commands, model configuration, API adapters, and SQLite state.
• tests/ checks provider routing, templates, and persisted switches without contacting external services.

Installing the ChatGPT plugin does not start a persistent Discord process. Host and run bot.app yourself to receive slash commands. Nothing includes provider keys or a Discord bot token.

Security and limits

Keep .env private, rotate exposed tokens, and grant the bot only the necessary server permissions. A customized base URL is operator-controlled and makes the bot send the selected prompt and optional credential to that server. Review that destination before configuring it. Provider requests have a 25-second timeout. The bot returns up to 1,800 characters in /ask and does not maintain conversation history. It uses an exact model ID supplied by the operator, so provider availability and charges depend on that account.

Licensed under Apache-2.0.