Skip to main content

​
Introduction

Freyavoice AI is built on several core concepts that work together to create powerful voice AI applications. Understanding these concepts is essential for building effective solutions.

​
Agents

Agents are the heart of Freyavoice AI. They are intelligent voice assistants powered by advanced AI models that can understand natural language, maintain context, and execute actions.

​
What Makes an Agent

An agent consists of several key components: System Prompt: The agent’s instructions and personality. This defines how the agent behaves, what it can do, and how it should respond to different situations. A well-crafted system prompt is crucial for agent performance. Model Configuration: The AI model that powers the agent (GPT-4, Claude, etc.). Different models have different strengths - some are better at reasoning, others at speed or cost efficiency. Voice Settings: How the agent sounds. This includes the voice provider, specific voice selection, speaking speed, and voice stability. The right voice can significantly impact user experience. Functions: Actions the agent can take. Functions allow agents to interact with external systems, query databases, make API calls, and perform other operations beyond just conversation. Context Management: How the agent maintains conversation history and context. This includes conversation length limits, context window management, and memory strategies.

​
Agent Lifecycle

When a call comes in, the agent goes through several stages:
  1. Initialization: The agent loads its configuration, system prompt, and any required functions
  2. Context Loading: Previous conversation history is loaded if available
  3. Listening: The agent receives and processes audio input
  4. Understanding: Speech is converted to text and analyzed
  5. Reasoning: The AI model processes the input and generates a response
  6. Action Execution: Any required functions are called
  7. Response Generation: The response is converted to speech and delivered
  8. Context Update: The conversation is saved for future reference

​
Workflows

Workflows are visual, node-based systems that allow you to create complex call routing and processing logic. They enable you to build sophisticated applications that go beyond what a single agent can do.

​
Workflow Components

Nodes: The building blocks of workflows. Each node represents a specific action or decision point:
  • Agent Nodes: Connect to AI agents to handle conversations
  • Function Nodes: Execute custom functions or API calls
  • Condition Nodes: Route calls based on conditions (time of day, caller ID, etc.)
  • Transfer Nodes: Transfer to human agents or other systems
  • Wait Nodes: Pause execution for a specified duration
  • Loop Nodes: Repeat actions based on conditions
Connections: The lines between nodes that define the flow of execution. Connections can be conditional, allowing different paths based on node outputs. Variables: Data that flows between nodes. Variables can store user input, function results, or any other data needed during call processing.

​
Workflow Execution

When a workflow is triggered:
  1. Entry Point: Execution starts at the initial node
  2. Node Processing: Each node executes its logic
  3. Data Flow: Variables are passed between nodes
  4. Conditional Routing: Conditions determine which path to take
  5. Parallel Execution: Some nodes can execute simultaneously
  6. Completion: The workflow completes when reaching an end node or transfer

​
When to Use Workflows vs Agents

Use a single agent when:
  • Your use case is straightforward
  • You need a simple conversation flow
  • You want to get started quickly
  • The agent can handle everything independently
Use workflows when:
  • You need complex routing logic
  • Multiple agents need to work together
  • You require conditional branching
  • You need to integrate with multiple systems
  • You want to add human-in-the-loop capabilities

​
Telephony Layer

The telephony layer connects your agents and workflows to the real world through phone numbers and call handling infrastructure.

​
Phone Numbers

Phone numbers are your connection point to the telephony network. They can be:
  • Imported: Brought in from providers like Twilio
  • Purchased: Acquired through Freya’s marketplace (if available)
  • Configured: Set up with specific capabilities (voice, SMS, etc.)

​
Call Routing

When a call comes in, Freya routes it through several stages:
  1. Inbound Reception: The call is received at the phone number
  2. Number Resolution: Freya identifies which agent or workflow to use
  3. Session Creation: A call session is created with unique identifiers
  4. Agent/Workflow Invocation: The configured agent or workflow is activated
  5. Call Processing: The agent/workflow handles the conversation
  6. Call Completion: The call ends and data is saved

​
Outbound Calling

For outbound calls, the process is similar but initiated by your application:
  1. Call Request: Your application requests an outbound call
  2. Number Selection: Freya selects the appropriate phone number
  3. Call Initiation: The call is placed to the destination
  4. Connection: Once connected, the agent/workflow takes over
  5. Processing: Same as inbound calls from this point

​
Integration Points

Freyavoice AI integrates with various external systems: Telephony Providers: Twilio, and other SIP providers for call handling AI Models: OpenAI, Anthropic, and other LLM providers for agent intelligence External APIs: Webhooks and function calls to integrate with your systems Storage: Conversation history, call recordings, and analytics data

​
Data Flow

Understanding how data flows through Freyavoice AI helps you build better applications:
  1. Audio Input: Phone call audio is received
  2. Speech-to-Text: Audio is converted to text
  3. Agent Processing: The agent analyzes the text and generates a response
  4. Function Execution: Any required functions are called
  5. Text-to-Speech: The response is converted back to audio
  6. Audio Output: The audio is sent back to the caller
  7. Data Storage: All interactions are logged for analytics and improvement

​
Next Steps

Architecture Deep Dive

Learn how these core concepts come together in Freyavoice AI’s architecture.