Fixie.ai operates as Ultravox, a speech-native voice AI platform built for developers creating conversational agents that process audio directly without text transcription. This system centers on Ultravox v0.7, a core model achieving 91.8% on Big Bench Audio benchmarks and 97% with reasoning enabled, paired with UltraVAD v0.1 for dynamic turn-taking prediction. Unlike systems that convert speech to text and back, this architecture preserves paralinguistic signals like tone, cadence, and pitch throughout interactions. The company manages the complete inference stack to minimize latency, offering REST APIs, web and mobile SDKs, and telephony integrations with major providers. Open weight models are available on Hugging Face. Pricing starts at 5¢ per minute including text-to-speech for up to five concurrent calls, with a Pay Go plan at $0 monthly, Pro at $100 monthly when billed yearly, and custom Enterprise options.
Development teams building voice AI products form the core audience, from startup engineers prototyping conversational interfaces to enterprise developers deploying at scale. This service suits software teams comfortable with API integration who need real-time voice interaction capabilities without managing complex audio processing infrastructure. Companies running customer service operations, healthcare applications requiring natural dialogue, or any organization embedding voice agents into products benefit from the managed inference stack. Both technical founders experimenting with voice AI and established engineering teams scaling existing voice systems find the developer-friendly tooling and SDK support accessible.
Typical implementations include customer support lines where agents handle inbound calls through telephony integrations, telehealth platforms conducting patient intake interviews with natural conversation flow, and mobile applications requiring voice-first interfaces. The speech-native architecture fits scenarios where preserving vocal nuance matters—coaching apps analyzing tone during practice sessions, mental health tools detecting emotional cues, or sales training platforms evaluating pitch and cadence. Development workflows involve integrating the REST API into existing applications, deploying SDKs for web or mobile channels, and connecting telephony providers for phone-based interactions. Teams building voice assistants for specific industries use this system's ability to maintain low latency while processing audio end-to-end, avoiding the delays inherent in transcription-based systems.
