Hathora
Explore, test & deploy production ready voice models.
201 followers
Explore, test & deploy production ready voice models.
201 followers
Build voice agents on open source or closed models with zero DevOps. Start instantly on shared endpoints and upgrade to dedicated infrastructure for privacy, compliance, or VPC requirements. Models run in 14 regions for ultra low latency. Bring your own models or custom containers as you scale.
Hathora
Hey Product Hunt! 👋 I'm Sid, co-founder of Hathora, and I'm excited to share Models powered by Hathora with you today.
The Problem:
We've talked to dozens of developers building voice AI applications, and the story is always the same: they spend more time wrestling with infrastructure than improving their agents. Deploying voice models globally, managing latency, handling scale — it's a massive distraction from the actual product. On the other hand, if you use closed source models from teams like OpenAI, ElevenLabs, and Cartesia you pay 7x the price.
Our Solution:
Models allows you to host your own models one (or many) from our model marketplace and fine-tune them on our managed infrastructure. We handle the global distribution, the infrastructure complexity, and the scaling - you focus on building great voice experiences.
Why We Built This:
At Hathora, we've been building infrastructure for real-time applications (originally for game servers, where even 20ms matters). We saw the same patterns emerging in voice AI: developers need low latency, global distribution, and reliable infrastructure. Models brings that same infrastructure-as-a-service approach to voice models.
What Makes Models Different:
Deploy to edge locations worldwide and give your global user base sub-50ms network latency
Use open-source models (no vendor lock-in, full customization)
Fine-tune on your own data with managed compute
Zero DevOps: we handle the infrastructure and provide serverless endpoints
Current Status:
We're in beta with 10-15 early users, and we're opening up access today. We'd love to hear your feedback and learn about your voice AI use cases!
Special Offer:
Product Hunt users one month UNLIMITED USE on Hathora with code PHMODELS, and the offer expires December 31.
Thanks for checking us out! Happy to answer any questions. 🚀
Follow us on PH, X, and LinkedIn to stay up-to-date with our community initiatives, both IRL and online.
@dsiddharth Many congrats on the launch! The 7x cost difference vs closed models is wild, but believable. :)
Hathora
@dsiddharth @rohanrecommends truly shocking, but believable. Can't wait foro your feedback onthe platform.
Hathora
@rohanrecommends It's crazy how much margin (or inefficiencies) some of these voice providers have!
Prava
Congrats! :)
Are GPU costs fully bundled into the platform pricing or billed separately?
Hathora
@zerotox The GPU cost is fully bundled when you hit our models.hathora.dev API. You do have the option to deploy the models on a dedicated endpoint and in that cause you will pay for GPU consumption.
Are there any limitations on model size, architecture or framework? BTW, congrats on the launch :D
Hathora
@himani_sah1 No there are no limitations on model size, architecture, or framework. You can deploy any custom docker container on Hathora.
Hathora
Thanks for the early support, folks! We'd love to see your videos using the tool here - some really fun things can be spit out of the chaining demo we built. You can find it right at the top of the page on models.hathora.dev
Daily.co
Congratulations on the launch, @dsiddharth. Open weights models, globally accessible at very low latency, is a big deal! I'm really excited about this.
A quick note for people here interested in voice AI: @taruni from Hathora will be on the panel at the voice AI meetup tonight. Join us in person in SF or via livestream. Come hang out and talk to Taruni about low-latency networking, edge inference, and model selection.
https://luma.com/900s2m9f
Hathora
@dsiddharth @taruni @kwindla love it - can't beat an in-person event to celebrate!
Speech to Note
I've already started demoing the models, after attending the launch event at Dubai. Pretty amazing, I must say. What other open source models are you planning to bring in both STT and TTS space?
Hathora
@abhishek_ux thanks for being an early adopter Abhishek! We'll be continuing to add support for open source TTS and STT models. Feel free to add your requests on our website 😊
Nas.io
Do you provide guardrails or monitoring for hallucinations / unsafe outputs at the infra level?
Hathora
@nuseir_yassin1 no we don't have any specific guardrails on the models itself, we only support hosting models for inference.
Hathora
@nuseir_yassin1 @taruni happy to chat further about this, it's a deeper question but we do have partnerships with higher level AI safety companies, but that would be higher up the stack.