Technical Specifications Document: The Water Company AI

Version: 1.0

Date: 14, February 2025

1. System Architecture

1.1 Overview

The Water Company platform enables users to build and manage AI-driven organizations using Python-based AI agents, teams, and companies. The system supports local and cloud deployment, with sandboxed execution for security and scalability.

1.2 Core Components

1. AI Agents: Python files managing AI models, actions, and knowledge bases.
2. AI Teams: Python files orchestrating agent workflows via RabbitMQ for team chat.
3. AI Companies: Python files aggregating team data into a company-wide knowledge base and orchestrating agent workflows via RabbitMQ for team chat.
4. Server Infrastructure:
   • Sandbox Environment: Isolated execution (Docker containers) for user/AI company.
   • Database: PostgreSQL for storing chat logs, actions, and company metadata.
   • File Storage: Local/cloud storage for RAG documents (e.g., MinIO, AWS S3).

2. Agent Architecture

2.1 Agent Components

1. Specialized AI Model:
   • Purpose: Domain-specific decision-making (e.g., legal, finance, coding).
   • Implementation: RAG-enabled model, fine-tuned with user-provided documents.
2. Actions AI Model:
   • Purpose: Execute pre-defined actions (e.g., SendEmail(), AnalyzeData()).
   • Implementation: Action-optimized models (e.g., Qwen 2.5, Gemini 2.0 Pro).
3. Knowledge Bases:
   • Personal RAG: Agent-specific documents (e.g., job instructions).
   • Team RAG: Shared files, metrics, and goals.
   • Knowledge bases can be made by creating knowledge graphs bases on the data provided (As according to the SciAgents paper https://arxiv.org/abs/2409.05556)
4. Action Library:
   • Pre-built actions (e.g., PostOnSocial()) and custom actions (Python/APIs).

2.2 Execution Flow