Mystic

Mystic is designed to make it easy to deploy and scale Machine Learning models with serverless GPU inference. The service lets you deploy ML models in your own cloud accounts (Azure, AWS, GCP) or in Mystic's shared GPU cluster. That flexibility is designed to help you get the best price and performance.

Among other features, Mystic offers:

• Cloud Integration: Native support for AWS, Azure and GCP means you'll have a smooth experience in the cloud you're already using.
• Fast Inference: Support for multiple inference engines like vLLM, TensorRT and TGI means your models run as fast as possible.
• Cost Optimizations: Multiple options to save you money, including running on spot instances, parallelizing GPU usage and using cloud credits.
• Managed Kubernetes Platform: A managed Kubernetes environment in your own cloud, so you get high performance without the hassle.
• Open-Source Python Library: Easy deployment of AI models with a Python library and API.

Mystic's scaling is automated, with the service adjusting GPU usage based on the number of API calls your models are receiving. That means you don't have to worry about over- or under-provisioning, and you don't pay for resources you're not using. You can also fine-tune scaling and cooldown periods with APIs.

Pricing is based on per-second compute usage for CPU and GPU instances, with different rates depending on the specific hardware you need. The Serverless plan comes with a $20 free credit and no account, egress or storage fees. The Bring Your Own Cloud plan, for enterprise customers, charges a flat monthly rate for using your cloud compute credits with Mystic's software.

Mystic is geared for teams working on text, image and video, or audio, and is designed to be a good way to handle AI models. By hiding the infrastructure details, Mystic lets data scientists and engineers concentrate on model development, not infrastructure.

Published on June 14, 2024