Inworld AI

Inworld AI provides realtime AI models for building voice-driven, character-based applications. The stack includes text-to-speech, speech-to-text, and LLM routing, unified behind a single API that supports full-duplex voice conversations, meaning a user and an AI can talk over each other and interrupt naturally, similar to a real conversation. The text-to-speech system supports voice cloning from as little as 5 to 15 seconds of reference audio, and offers steering controls across eight dimensions of natural-language direction, such as tone or emotion. The speech-to-text system profiles voices for emotion, accent, and vocal style. The Realtime Router gives access to more than 220 large language models across third-party and first-party providers, including optimized open-source models like Gemma, DeepSeek, and GLM variants. The platform is built for developers creating consumer-facing applications: companions, social apps, games, customer support voice agents, sales agents, phone agents, language-learning tools, and interactive media. It exposes an OpenAI Realtime protocol-compatible WebSocket API along with REST endpoints, and offers cloud hosting with enterprise on-premise deployment options. Named customers include Wishroll, Bible Chat, and Talkpal. Inworld originally focused on AI game characters and NPCs before expanding into this broader realtime voice API product; the site does not state a specific founding year or public pricing.