Design Proposals

Note: This is the original design proposal created before implementation began. For current documentation, see the main docs sections.

Overview

BILL is a plugin-based AI agent that maintains consistent personality across Twitter and Telegram while keeping platform-specific conversations separate.

Key Features

• Plugin Architecture: Separate modules for each platform

• Intelligent LLM Routing: Automatic selection between Claude, GPT-4, and Llama based on task complexity

• Image Generation & Analysis: GPT-4o vision for image analysis and DALL-E 3 for image creation

• Hybrid Memory System: Platform-isolated conversations + shared knowledge with smart context selection

• Character Consistency: Same personality across platforms

• Simple Deployment: Managed services, no complex infrastructure

Technology Stack

• Runtime: Node.js + TypeScript + Bun

• Database: Supabase (PostgreSQL)

• Vector DB: Pinecone

• LLM: OpenRouter

• Image Gen: OpenAI GPT-4

• Deployment: Railway

MVP Scope

Included

✅ LLM Routing: Claude, GPT-4, Llama selection based on task ✅ Image Generation: DALL-E 3 image creation ✅ Image Analysis: GPT-4o vision capabilities ✅ Twitter Service ✅ Telegram bot (DMs and groups) ✅ Dual-layer memory system ✅ Character configuration ✅ Simple deployment

Not Included (Future)

❌ Load balancing & CDN ❌ Advanced analytics ❌ Auto-scaling

Quick Start

1. Setup Services: Supabase, Pinecone, OpenAI, Twitter API, Telegram Bot

2. Configure Environment: Copy .env.example and fill in API keys

3. Deploy: Connect to Railway/Render and deploy

4. Test: Send mentions on Twitter or message the Telegram bot

Documentation

• System Architecture - Core components and data flow

• Implementation Guide - Code structure and interfaces

• Configuration - Character setup and environment variables

• Deployment - Simple deployment guide

Cost Estimate

~$150-200/month for MVP (Pinecone being the largest cost at $70+/month)

---