Inworld

Inworld is a voice AI platform and agent runtime for developers and enterprises building real-time conversational applications such as AI companions, customer support agents, and voice-enabled experiences.

Key capabilities include:

• Real-time voice with a top-ranked text-to-speech engine (sub-200ms latency), plus speech-to-speech for controllable live conversations.
• Speech-to-text for realtime transcription, including streaming over WebSocket and features like word-level timestamps, diarization, and custom vocabulary.
• Model-agnostic orchestration with an LLM Router that routes across OpenAI, Anthropic, Google, Mistral, and 200+ models, with built-in failover, analytics, and A/B testing.
• Agent and character building via web-based Studio tools, including customizable personalities, emotions, goals, multimodal interactions (text, speech, gestures), and long-term memory, plus Contextual Mesh for custom knowledge and safety constraints.

Pricing: Freemium/ Usage-based.

• New users receive free credits on signup to explore the platform.
• Pay-per-use model with no markup on underlying LLM/STT/TTS provider costs. You purchase credits (minimum $10) and are billed based on consumption (e.g., per 1M characters for TTS, per second for STT, per million tokens for LLMs).
• On-Demand plan for developers/startups and Enterprise plan (with volume discounts, HIPAA, on-prem options, dedicated support, SLA, etc.) are available.

Notable for combining low-latency voice infrastructure with a production-focused orchestration layer (observability, experimentation, routing) and for originating in expressive character AI for games while expanding to enterprise voice agents and realtime apps at scale.