# Build AI Voice Agents in Minutes with Inworld AI

Source: https://www.youtube.com/watch?v=OSz9lD34lHI
Recap page: https://rapidrecap.app/video/OSz9lD34lHI
Generated: 2026-01-28T15:43:57.201+00:00

---
## Quick Overview

The video demonstrates building a functional AI voice agent companion in minutes using the Inworld AI API, specifically highlighting the new Model 1.5 which offers up to 25x cost savings compared to competitors and features sub-250ms latency for highly realistic human conversation.

**Key Points:**
- Inworld Model 1.5 is dirt cheap, costing between half a cent to only one cent per minute, making it around 25x cheaper than competitors, which historically limited voice-to-voice applications due to cost.
- The latency for conversation response is below 250 milliseconds, which is faster than typical human response times, creating an uncanny, real human conversation experience.
- The platform provides an open-source code template for a voice agent accessible via a GitHub link, enabling rapid development in cost-effective ways.
- The speaker successfully created a TypeScript web application companion named Olivia, customizing its personality using options like Supportive, Witty, Sarcastic, Wise Mentor, and Chill Friend.
- The development process leverages AI coding assistance by pasting Inworld's documentation references directly into a planning tool, which then generates the necessary TypeScript application structure.
- The technology is applicable across various verticals, including healthcare for prerequisite data collection, customer support, AI companions, fitness trainers, and notably, for creating contextual, memory-retaining NPCs in video games.
- The final application requires running both a server command (CD server and npm start) and a client command (client npm start) to establish WebSocket connection and run the frontend.

**Context:** The video serves as a sponsored tutorial showcasing the capabilities of the Inworld AI API, focusing on their newly released Model 1.5 for building advanced AI voice agents. The presenter, Olivia, guides the audience through leveraging this technology to quickly construct a personalized AI best friend capable of real-time, low-latency voice interaction, emphasizing the massive leaps in cost-effectiveness and conversational realism achieved by the new model.

## Detailed Analysis

The core value proposition of Inworld Model 1.5 lies in its unprecedented cost efficiency and speed; it costs only one cent per minute or less, drastically undercutting competitors by a factor of 25, thus removing a major barrier to deploying voice-to-voice AI applications. Furthermore, the model achieves sub-250ms latency, mimicking or exceeding human conversational speed, which is crucial for creating natural interactions. The tutorial walks through using Inworld's provided playground to preview various voices like Hades and Luna, and then immediately transitions to building a TypeScript web application using open-source templates accessible via GitHub documentation. The development process is significantly streamlined by using an AI coding assistant where the user pastes Inworld's documentation links, allowing the tool to generate a plan and boilerplate code for a Node.js environment, which can also be adapted for environments like Unreal or Unity for game NPCs. The creation involves generating an API key with full permissions, defining desired personalities (e.g., Chill Friend), and managing the dual frontend/backend execution environment using separate npm start commands for the client and server to establish the WebSocket connection. The resulting agent can dynamically adopt different personalities defined in system prompts, allowing for tailored interactions, such as a simple chat or a highly contextual customer service bot that could eventually process image inputs as well.

### Inworld Model 1.5 Key Features

- Cost-effective at 0.5 to 1 cent per minute
- Latency below 250ms for fast responses
- Supports text, speech, and image input capabilities

### Rapid Development Workflow

- Utilizes pre-made open-source code templates linked from documentation
- Leverages AI coding assistance by pasting API documentation for plan generation
- Requires generating and setting a fully-permissioned API key in an .env file

### Application Building Steps

- Create a TypeScript app leveraging Inworld speech input/output
- Select desired personalities (Supportive, Witty, Chill Friend, etc.)
- Run separate commands for server (backend/websocket) and client (frontend) execution

### Use Cases and Potential

- Highly beneficial for healthcare data collection and customer support
- Future use in video games for NPCs that contextually remember past conversations
- Demonstrated creation of a personalized AI companion named Olivia

### Personality Customization

- Agents adopt distinct behaviors based on system prompts within the personality file
- System prompts allow for detailed instruction, such as requesting plain text output instead of asterisks
- Users can easily switch between voices (e.g., Olivia, Hades) and conversational styles.