Hey HN, this is Harsh from Hathora.

We started as a game server hosting company and are launching a new offering today centered around voice models, leveraging our existing platform that runs game servers.

We hope to achieve a few things with this product: 1. Democratized access to a wide variety of open source and licensed voice models 2. Ability to mix and match speech to text, reasoning (llm), and text to speech models to form inference chains for agentic voice workflows 3. Focus on latency reduction via (a) running models in 14+ regions around the globe, and (b) colocating models in the same datacenter to reduce networking overhead within a inference chain

The models are made accessible as Serverless API endpoints with token/duration based billing, and also available as dedicated deployments for enterprises with specific compliance requirements.

You can check it out here: https://models.hathora.dev/