Build AI Voice Agents in Minutes with Inworld AI

Creating an AI Companion with InWorld 1.5

Key Concepts:

• InWorld 1.5: A new AI model specializing in text-to-speech, speech-to-text, and multimodal input (text, speech, image).
• Latency: The delay between input and response in a conversation (InWorld 1.5 boasts <250ms latency).
• LLM (Large Language Model): The underlying technology powering the conversational abilities of the AI.
• API (Application Programming Interface): A set of rules and specifications that allows different software applications to communicate with each other.
• Websocket: A communication protocol enabling real-time, two-way communication between a client and a server.
• GitHub: A platform for version control and collaborative software development, used here to access open-source code templates.
• TypeScript: A superset of JavaScript that adds optional static typing.
• ENV File: A file used to store environment variables, such as API keys, for security and configuration.

---

I. Introduction & InWorld 1.5 Overview

The video demonstrates building a personalized AI companion using InWorld 1.5, a recently launched AI model. The presenter, Olivia, highlights InWorld’s capabilities in handling various inputs (text, speech, image) and its significant advantages over competitors. A key focus is the model’s cost-effectiveness and low latency, enabling realistic and responsive conversations. The video is sponsored by InWorld.

II. Cost & Performance Advantages

InWorld 1.5 is presented as a game-changer due to its pricing. The cost is between half a cent and one cent per minute of speech, approximately 25 times cheaper than competing models. This drastically reduces the financial barrier to entry for developing voice-based applications. Beyond cost, the model’s low latency (under 250 milliseconds) is emphasized. This speed is faster than typical human response times, creating a more natural conversational flow. Olivia notes that high costs have historically hindered the development of voice-to-voice applications, a problem InWorld 1.5 addresses.

III. Exploring the InWorld Playground & Templates

InWorld provides a playground environment for experimentation. Users can select from a diverse range of pre-built voices (e.g., Hades, Luna, Giani Italiano) to customize their AI companion. The platform also offers pre-made, open-source code templates accessible via GitHub, specifically a voice agent template. This allows developers to quickly prototype applications without starting from scratch. The presenter highlights the documentation available alongside the code.

IV. Potential Use Cases

Several potential applications of InWorld’s technology are discussed:

• Healthcare: Automating initial data collection during phone calls, streamlining patient access to information.
• Customer Support: Providing efficient and cost-effective support interactions.
• AI Companion: Creating personalized conversational partners (the focus of the video).
• Fitness Trainer: Developing interactive fitness coaching experiences.
• Social Media: Building engaging AI personalities for social platforms.
• Video Games (NPCs): Creating Non-Player Characters (NPCs) with dynamic, context-aware dialogue and memory of past interactions with the player. This is presented as a potential future direction for more immersive gaming experiences.

V. Building the AI Companion Application – Step-by-Step

The video details the process of building a basic AI companion application using InWorld 1.5 and a code generation tool (Cursor).

1. Project Setup: A new TypeScript application is created using Cursor, specifying the need for speech input/output and a personality selection dropdown.
2. API Key Generation & Configuration: An InWorld API key is generated, granting full permissions to access all functionalities. This key is then stored in an environment (.env) file for security.
3. Code Generation with Plan Mode: Cursor’s “plan mode” is used to generate the application code based on the provided specifications and InWorld API documentation. The generated code is presented, and the presenter encourages viewers to review and modify it.
4. Running the Application: The application is launched using npm start for both the frontend and backend. Troubleshooting steps are shown, including resolving port conflicts.
5. Testing & Customization: The application is tested with sample prompts ("Do you like the beach or the mountains?"). The presenter demonstrates how to adjust the AI’s responses by modifying the system prompt and personality settings. Specifically, they show how to remove unwanted characters (like asterisks) from the output.

VI. Technical Details & Code Exploration

The presenter briefly explores the generated code, highlighting key elements:

• Voice ID: Used to identify the selected voice for the AI companion.
• System Prompt: A crucial component that defines the AI’s personality, behavior, and context. The presenter emphasizes the importance of tailoring the system prompt to the specific application (e.g., healthcare, customer service).
• Multimodal Input Potential: The presenter notes the potential to extend the application to handle image input, allowing the AI to analyze visual information (e.g., screenshots from customer support requests).

VII. Notable Quotes

• “This is dirt cheap for the quality we're getting.” – Olivia, regarding the cost of InWorld 1.5.
• “It's going to take me more than 250 milliseconds to respond. So, let's build an app.” – Olivia, emphasizing the low latency of the model.
• “This is truly next level stuff.” – Olivia, commenting on the ease of use and power of the InWorld API.

VIII. Conclusion & Takeaways

The video successfully demonstrates the ease and affordability of creating a functional AI companion using InWorld 1.5. The key takeaways are:

• Accessibility: InWorld 1.5 significantly lowers the barrier to entry for developing AI-powered voice applications.
• Realism: The model’s low latency and advanced conversational capabilities create a more natural and engaging user experience.
• Versatility: The platform supports a wide range of use cases, from healthcare and customer support to gaming and personal companionship.
• Developer-Friendly: The open-source templates, comprehensive documentation, and code generation tools simplify the development process.
• Future Potential: The ability to handle multimodal input (text, speech, image) opens up exciting possibilities for more sophisticated AI applications.

The presenter concludes by encouraging viewers to explore the InWorld API and highlighting the transformative potential of this new technology.