System Requirements Document (SRD)

steel-agent

1. Introduction

The steel-agent project is a unified AI application designed to house multiple specialized agents, each excelling in a specific domain. These agents are pre-trained with expertise in areas such as emotional support, productivity enhancement, and real-time news aggregation. The app is designed to provide a seamless, engaging, and user-friendly experience, where users can interact with these agents individually or benefit from their collaborative insights.

This document outlines the system requirements for steel-agent, incorporating Krishna's vision of an engaging, cross-connected, and conversational AI ecosystem.

2. System Overview

The steel-agent app is a single platform that integrates multiple AI agents, each with pre-built expertise in specific domains. The agents are designed to actively engage users, collaborate with each other, and deliver insights in a friendly, conversational manner.

Key features include:

• Emotion Support Agent: Provides empathetic and supportive interactions to improve emotional well-being.
• Productivity Agent: Offers personalized tips and strategies to enhance daily productivity.
• News Aggregation Agent: Collects and summarizes the latest internet news and updates, delivering them to platforms like Slack.

The app emphasizes user engagement, cross-agent collaboration, and a conversational interface to ensure a seamless and enjoyable experience.

3. Functional Requirements

• As a User, I should be able to interact with each agent in a conversational and friendly manner.
• As a User, I should receive proactive engagement from agents, such as check-ins or suggestions.
• As a User, I should be able to receive news summaries as text updates in Slack or similar platforms.
• As a User, I should benefit from cross-agent collaboration, such as the productivity agent adjusting recommendations based on insights from the emotion support agent.
• As a User, I should experience a unified interface where all agents are accessible within a single app.
• As a User, I should feel that the agents are approachable, engaging, and human-like in their tone and interactions.

4. User Personas

4.1 General User

• Description: Individuals seeking personalized AI assistance for emotional support, productivity, and news updates.
• Goals:
  • Improve emotional well-being through empathetic interactions.
  • Enhance daily productivity with actionable tips.
  • Stay informed with concise news summaries.
• Pain Points:
  • Overwhelming information overload.
  • Lack of personalized support for emotional and productivity needs.

5. Visuals Colors and Theme

To reflect the engaging and friendly nature of the steel-agent app, the following unique color palette has been designed:

• primary: #4A90E2 (Vibrant Sky Blue)
• primary_light: #A6C8F0 (Soft Sky Blue Tint)
• secondary: #F5A623 (Warm Amber)
• accent: #D0021B (Bright Crimson Red)
• highlight: #F8E71C (Golden Yellow)
• bg: #F7F9FC (Light Grayish Blue)
• surface: rgba(255, 255, 255, 0.9) (Soft White)
• text: #333333 (Dark Charcoal)
• text_muted: #7F8C8D (Muted Gray)
• border: rgba(0, 0, 0, 0.1) (Subtle Black Tint)

This palette ensures a professional yet approachable aesthetic, with vibrant accents to highlight key interactions.

6. Signature Design Concept

"Interactive Agent Universe"

The home page of the steel-agent app will feature an interactive galaxy map where each agent is represented as a glowing, animated star. Users can:

• Hover over a star to see a brief description of the agent's expertise.
• Click on a star to "zoom in" and enter a dedicated conversational space for that agent.
• Drag the galaxy map to explore other agents, with smooth transitions and physics-based animations.

The galaxy will feature subtle parallax effects, with distant stars and constellations moving at different speeds as users navigate. The background will shift colors dynamically based on the time of day (e.g., soft blues in the morning, warm oranges in the evening).

This concept will be built using @react-three/fiber and @react-three/drei for the 3D galaxy, with gsap for smooth animations and transitions.

7. Interaction Model & Motion Direction

Interaction Model: Parallax

The landing page will use a parallax interaction model, creating a sense of depth and immersion. Layers of stars, constellations, and agent icons will move at varying speeds as users scroll or drag the galaxy map.

Motion Direction: Animated

• Hover Effects: Stars will glow and expand slightly when hovered over.
• Click Transitions: Smooth zoom-in animations when entering an agent's conversational space.
• Micro-Interactions: Subtle animations, such as twinkling stars and pulsating highlights, will enhance the user experience.

Internal pages (e.g., agent chat interfaces) will prioritize clarity and simplicity, with minimal motion to maintain focus on the conversation.

8. Non-Functional Requirements

• The app should support seamless integration with Slack for delivering news summaries.
• The app should maintain a response time of less than 2 seconds for all interactions.
• The app should be accessible across devices, including desktops, tablets, and smartphones.
• The app should ensure data privacy and security, adhering to GDPR and other relevant regulations.

9. Tech Stack

Frontend

• React for Web

Backend

• Python
• FastAPI

Database

• MySQL or MariaDB (with Alembic for migrations)

AI Models

• GPT 5.4 for user-friendly responses
• Claude 4.6 Opas for academic or coding work
• Gemini 3.1 Pro for friendly interactions

AI Tools

• Litellm for LLM Routing
• Langchain

Orchestration

• Docker for local orchestration
• Kubernetes for server-side orchestration

10. Assumptions and Constraints

• The app assumes users have access to Slack or similar platforms for receiving news updates.
• The app will rely on pre-trained AI models and will not support user-driven training.
• The app will require an active internet connection for all interactions.

11. Glossary

• Agent: A specialized AI entity with expertise in a specific domain.
• Parallax: A design technique where background and foreground elements move at different speeds to create depth.
• LLM: Large Language Model, used for generating human-like text responses.