Contact

Tell us what you serve and where it runs

Questions about setup, pricing, security, or whether Sirvich fits your stack. An infrastructure engineer reads every message and replies within one business day.

Talk to an engineer

Every message reaches someone who runs GPUs for a living

Send the shape of your workload: the models, the traffic, and where it runs today. We will tell you plainly whether we can help.

Before you write

Answers to what teams ask first

Does Sirvich run in our cloud account?: Yes, on the Scale and Enterprise plans. Replicas run on GPUs in your AWS, Google Cloud, or Azure account, or on your own Kubernetes cluster. Hobby and Team use our shared GPUs.
Which models can we deploy?: Any model that runs in a container. Sirvich has tuned runtimes for open language models, embeddings, speech, and vision, and you can bring your own server.
What happens when a rollout goes wrong?: Every new version starts as a canary on a small share of traffic. If latency, errors, or your evals get worse, Sirvich rolls back and tells you which check failed.
How is GPU time billed?: By the second, for the time a replica is running. Idle replicas scale down, so a quiet night costs close to nothing.
How long does setup take?: Most teams serve their first model in an afternoon: connect a cloud account, point us at a model, and set a latency target.