System Requirements Document (SRD) for Onyx-Tenant

1. Introduction

The Onyx-Tenant project is a multi-tenant customer support system designed to empower tenants with advanced voice agent capabilities and intuitive workflow customization. Built on top of the Ten Framework, this system will provide a robust platform for tenants to manage their customer support operations efficiently. The system will integrate both third-party and proprietary LLM (Large Language Model) solutions, enabling seamless interaction through text and voice interfaces.

This document outlines the system requirements for Onyx-Tenant, ensuring alignment with the project's goals and John Guo's vision for a scalable, user-friendly, and feature-rich platform.

2. System Overview

The Onyx-Tenant system will cater to three primary user groups:

1. End Users: Customers of the tenants who interact with the system via text-based questions or phone calls (SIP support).
2. Tenants: Businesses subscribing to the platform, who can self-manage their customer support workflows, deploy agents, and monitor usage, billing, and agent health.
3. System Operators: Administrators responsible for managing tenants, allocating compute resources, monitoring model usage, and generating invoices.

The system will leverage a combination of third-party and proprietary ASR (Automatic Speech Recognition), TTS (Text-to-Speech), and LLM models to deliver high-quality voice agent functionality. A visual interface for workflow graph creation will make it easy for tenants to define and customize their customer support processes.

3. Functional Requirements

• End User Functionalities:

  • As an End User, I should be able to ask questions via a text-based UI.
  • As an End User, I should be able to interact with the system via phone calls using SIP support.

• Tenant Functionalities:

  • As a Tenant, I should be able to self-subscribe to the platform.
  • As a Tenant, I should be able to define workflow graphs through a visual drag-and-drop interface.
  • As a Tenant, I should be able to deploy voice agents for customer support.
  • As a Tenant, I should be able to monitor the usage, billing, and health status of my agents.

• System Operator Functionalities:

  • As a System Operator, I should be able to disable a tenant's account.
  • As a System Operator, I should be able to allocate compute resources for tenants' agents.
  • As a System Operator, I should be able to monitor the usage of ASR, LLM, and TTS models.
  • As a System Operator, I should be able to create and manage tenant invoices.

4. User Personas

4.1 End User

• Description: Customers of the tenants who interact with the system for support.
• Goals: Resolve queries quickly via text or voice.
• Technical Proficiency: Low to moderate.

4.2 Tenant

• Description: Businesses subscribing to the platform to manage their customer support operations.
• Goals: Customize workflows, deploy agents, and monitor performance.
• Technical Proficiency: Moderate to high.

4.3 System Operator

• Description: Administrators managing the platform's tenants and resources.
• Goals: Ensure smooth operation, manage resources, and oversee tenant activities.
• Technical Proficiency: High.

5. Visuals Colors and Theme

The Onyx-Tenant platform will feature a modern, professional, and approachable design. The color palette is tailored to evoke trust, innovation, and clarity.

• Background: #F4F7FA (soft light gray)
• Surface: #FFFFFF (pure white)
• Text: #2D3748 (dark slate gray)
• Accent: #4A90E2 (vivid blue)
• Muted: #A0AEC0 (cool gray)

This palette ensures readability and a clean, professional aesthetic while maintaining a touch of vibrancy through the accent color.

6. Signature Design Concept

The Onyx-Tenant homepage will feature an interactive 3D workflow graph as its centerpiece. This bold and memorable design will immediately immerse users in the platform's core functionality.

Key Features:

1. 3D Workflow Graph:

   • The homepage will display a dynamic, interactive 3D graph representing a sample customer support workflow.
   • Nodes in the graph will represent different stages of the workflow (e.g., "Customer Query," "Agent Response," "Escalation").
   • Users can click on nodes to see animations and tooltips explaining their functionality.

2. Interactive Animations:

   • The graph will rotate and zoom as users hover over or click on nodes, creating a sense of depth and engagement.
   • Smooth transitions will guide users through the workflow, making the system's capabilities intuitive and visually appealing.

3. Color Transitions:

   • The graph's nodes will adopt the platform's accent color (#4A90E2) when active, while inactive nodes will use the muted tone (#A0AEC0).
   • Background gradients will subtly shift based on user interactions, adding a dynamic and modern feel.

4. Micro-Interactions:

   • Hover effects, subtle vibrations, and glowing outlines will provide feedback as users explore the graph.

This design concept will not only showcase the system's visual workflow capabilities but also leave a lasting impression on users, setting Onyx-Tenant apart from competitors.

7. Non-Functional Requirements

• Scalability: The system must support up to 10,000 tenants and 1 million end users concurrently.
• Performance: Response times for text and voice interactions must not exceed 1 second under normal load.
• Reliability: The system must maintain 99.9% uptime.
• Security: All data must be encrypted in transit and at rest. Multi-factor authentication (MFA) will be required for tenant and system operator accounts.
• Compliance: The system must comply with US regulations, including GDPR and CCPA where applicable.

8. Tech Stack

Frontend

• Web: React
• Mobile: React Native

Backend

• Language: Python
• Framework: FastAPI

Database

• RDBMS: MySQL (with Alembic for migrations)
• NoSQL: MongoDB
• VectorDB: WeaviateDB

AI Models

• Third-Party: GPT 5.2, Claude 4.5 Opas, Gemini 3 Pro
• Proprietary: Custom LLM models for tenant-specific workflows
• Image Generation: Google Nano Banana

AI Tools

• LLM Routing: Litellm
• Workflow Orchestration: Langchain

Orchestration

• Local: Docker, docker-compose
• Server-Side: Kubernetes

9. Assumptions and Constraints

• Tenants will have varying levels of technical expertise, so the visual workflow graph must be intuitive and easy to use.
• The system will rely on both third-party and proprietary AI models, requiring robust integration and fallback mechanisms.
• SIP support for phone calls will be implemented using industry-standard protocols.
• The system will initially target the US market, with potential for global expansion.

10. Glossary

• ASR: Automatic Speech Recognition.
• TTS: Text-to-Speech.
• LLM: Large Language Model.
• SIP: Session Initiation Protocol, used for managing voice calls.
• Tenant: A business subscribing to the Onyx-Tenant platform.
• End User: A customer of a tenant interacting with the system.
• System Operator: An administrator managing the platform's tenants and resources.