AGENTS.md

AGENTS.md

This file provides guidance to LLM-based coding agents when working with code in this repository.

Project Overview

FediChatBot is an LLM-powered chatbot for the fediverse, built with Deno. It uses BotKit (a framework for building fediverse bots on top of Fedify) and Google's Gemini 3 Flash model via LangChain.

Development Commands

# Run development server with hot reload
deno task dev

# Type check, lint, and format check
deno task check

# Individual checks
deno check *.ts      # Type checking
deno lint            # Linting
deno fmt --check     # Format check
deno fmt             # Auto-format

Architecture

The entire bot is contained in a single file bot.ts (~350 lines). Key components:

• Bot Setup: Uses @fedify/botkit to create a fediverse bot with Deno KV for persistence
• LLM Integration: Uses LangChain with ChatGoogleGenerativeAI (Gemini 3 Flash)
• Event Handlers:
  • bot.onFollow: Sends greeting when someone follows
  • bot.onMention: Responds to mentions (not replies)
  • bot.onReply: Handles threaded conversations with context
• Prompt Templates: Located in prompts/ directory (system.txt, follow.txt, mention.txt, reaction.txt)
• Image Handling: Caches images in Deno KV with 3-hour expiry, converts to data URLs for LLM

Key Dependencies

• @fedify/botkit: Fediverse bot framework (docs: https://botkit.fedify.dev/llms.txt)
• @fedify/fedify: ActivityPub federation library (docs: https://fedify.dev/llms.txt)
• @langchain/google-genai: Google Gemini integration
• @logtape/logtape: Structured logging
• tinyld: Language detection for responses

Environment Variables

Requires Google AI API key for Gemini access (loaded via @std/dotenv).

License

AGPL-3.0-only