Xybrid is the operating layer for on-device AI that scales. It provides a unified API for any backend, enabling models to be shipped over the air and run on any platform and fallback to the cloud when necessary.
It's a privacy-first, cost-optimized solution for running ML models, supporting any model on any platform.
Hybrid Execution: Models can run on-device or in the cloud with intelligent routing based on factors like model size, available memory, battery life, and latency budget.
Zero Infrastructure: Eliminates the need for managing complex cloud infrastructure.
Privacy-First: Data stays on-device by default, with telemetry being opt-in and aggregated.
Cost-Optimized: Significantly reduces inference costs compared to cloud-only solutions, with on-device execution being free.
Broad Platform Support: Offers native SDKs for iOS, macOS, watchOS, tvOS, visionOS, Android, Flutter, Unity, Linux/Edge, and more.
Over-the-Air (OTA) Model Updates: Deploy new models without code changes, with options for canary, region-scoped, or fleet-wide rollouts.
Fleet Operations: A console interface allows for operating and monitoring AI model fleets at scale, including telemetry, device management, and model registry.
Extensive Model Support: Built for leading open-weight models, with day-0 support for new state-of-the-art models and automated artifact generation for various mobile and edge platforms.
Key features include:
One API for Every Backend: Seamlessly integrate AI models across diverse platforms.
Hybrid Routing: Dynamically route requests between on-device and cloud execution.
OTA Model Updates: Effortlessly update models without app store reviews.
Evals Across Runtimes: Compare model performance across different hardware and OS versions.
Per-Model Optimization: Automatically tune prompts for optimal performance on various backends.
The Xybrid runtime is open source, providing flexibility and transparency. The platform offers additional features for fleet management and operations, with a generous free tier for smaller teams.
Built with