GPT Discord

Integrates OpenAI's GPT models directly into Discord's message interface using discord.py's event handlers and cog architecture. Maintains per-user and per-channel conversation histories in memory or persistent storage, automatically handling Discord's message length limits (2000 chars) by splitting long responses across multiple messages. Uses a conversation state machine to track context across turns, enabling coherent multi-message exchanges within Discord's native threading and reply system.

Wraps OpenAI's DALL-E API (DrawDallEService cog) to generate images from text prompts within Discord. Handles image size/quality parameters, downloads generated images, and uploads them as Discord attachments with automatic fallback to URL embeds if upload fails. Supports prompt engineering via system instructions and integrates with the conversation context to generate images based on prior discussion.

Uses discord.py's interaction deferral mechanism to handle long-running operations (image generation, web search, code execution) without triggering Discord's 3-second interaction timeout. Defers the interaction immediately, then edits the response once the operation completes. Supports background task queuing for operations that exceed Discord's timeout window, with status updates via message edits or follow-up messages. Implements exponential backoff for API retries and graceful error handling.

Centralizes bot configuration via environment variables (API keys, Discord token, database URLs) and per-server settings stored in Discord (via guild-specific configuration channels or database). Supports feature flags to enable/disable capabilities per server, custom system prompts per channel, and role-based feature access. Uses Python's dotenv for local development and environment-based configuration for production deployment. Implements configuration validation and defaults for missing settings.

IndexService cog creates embeddings from documents (PDFs, websites, text) using OpenAI's embedding API, stores them in Pinecone or Qdrant vector databases, and enables semantic search via cosine similarity. Supports bulk indexing of websites via web scraping, document chunking with configurable overlap, and namespace isolation per user/server. Integrates with conversation context to inject relevant document snippets as RAG (Retrieval-Augmented Generation) context before sending queries to GPT.

SearchService cog integrates web search APIs (Google Custom Search, Bing, or similar) to fetch real-time information from the internet. Parses search results, extracts relevant snippets, and injects them into GPT context as grounding data. Supports follow-up searches based on conversation context and caches results to reduce API calls. Enables the bot to answer questions about current events, recent news, and real-time data that would be outside its training data cutoff.

CodeInterpreterService cog executes Python code in isolated environments (using exec() with restricted globals/locals or containerized execution) and returns stdout/stderr output. Supports multi-line code blocks, variable persistence across code cells within a session, and visualization output (matplotlib, plotly). Integrates with conversation context to execute code snippets discussed in chat and display results inline.

TranslationService cog uses DeepL, Google Translate, or OpenAI's translation capabilities to translate text between 100+ language pairs. Supports bulk translation of conversation history, maintains glossaries for domain-specific terminology, and preserves formatting (code blocks, mentions, emojis). Integrates with conversation context to translate previous messages or entire threads, enabling cross-language communication in multilingual Discord servers.

TranscribeService cog integrates OpenAI's Whisper API or similar speech-to-text services to transcribe audio files uploaded to Discord. Supports speaker diarization (identifying different speakers), timestamp alignment for long audio, and automatic language detection. Handles Discord audio file formats, downloads attachments, sends to transcription API, and returns timestamped transcripts with optional speaker labels. Integrates with conversation context to make transcripts searchable and indexable.

ModerationsService cog uses OpenAI's Moderation API or custom ML models to flag potentially harmful content (hate speech, violence, sexual content, etc.) in Discord messages. Supports configurable severity thresholds, per-server policy customization, and action automation (delete, warn, mute, ban). Integrates with Discord's audit log and can trigger notifications to moderators. Maintains moderation statistics and can generate reports on policy violations.

Model abstraction layer supports multiple LLM providers (OpenAI GPT-3.5/4, Anthropic Claude, open-source models via Ollama) with dynamic switching based on cost, latency, or availability. Implements provider fallback logic — if OpenAI is rate-limited, automatically routes to Claude or local Ollama. Supports different model capabilities (vision, function calling, long context) and automatically selects appropriate model for task. Configuration-driven provider selection enables cost optimization without code changes.

Implements conversation namespace isolation using Discord user IDs and channel IDs as keys, storing separate conversation histories and context for each user/channel combination. Integrates with Discord's role system to enforce access control — users can only access conversations they have permission to view. Supports shared conversation contexts for team channels while maintaining privacy for DMs. Uses Discord's permission system to determine visibility and edit rights.