Xybrid empowers developers to integrate on-device Artificial Intelligence capabilities, including Large Language Models (LLMs), Automatic Speech Recognition (ASR), and Text-to-Speech (TTS), directly into applications and games. It targets mobile, desktop, and game development platforms, offering a significant benefit of enhanced user privacy and offline functionality by eliminating the need for cloud-based AI processing.

How It Works

Xybrid is built upon a core Rust runtime, providing native bindings for a wide array of platforms such as Flutter, Unity, Swift, Kotlin, and a command-line interface (CLI). This unified Rust core ensures consistent model support and behavior across all integrated SDKs. The architecture prioritizes on-device inference, allowing AI models to run locally, which is advantageous for privacy-sensitive applications and environments with intermittent or no internet connectivity.

Quick Start & Requirements

Installation varies by platform:

• CLI: curl -sSL https://raw.githubusercontent.com/xybrid-ai/xybrid/master/install.sh | sh (macOS/Linux) or irm https://raw.githubusercontent.com/xybrid-ai/xybrid/master/install.ps1 | iex (Windows).
• Flutter: Add xybrid_flutter: ^0.2.1 to pubspec.yaml.
• Kotlin: Add implementation("ai.xybrid:xybrid-kotlin:0.2.1") to build.gradle.kts.
• Swift: Use Swift Package Manager with the URL https://github.com/xybrid-ai/xybrid.git.
• Unity: Install via Unity Package Manager using https://github.com/xybrid-ai/xybrid.git#upm.
• Rust: Add xybrid = "0.2.1" to Cargo.toml.

Official documentation is available at docs.xybrid.dev.

Highlighted Details

• Privacy-First: All AI inference occurs on-device, ensuring user data remains private.
• Offline Capable: Applications function without an internet connection after initial model downloads.
• Cross-Platform: A single API provides access across iOS, Android, macOS, Linux, and Windows.
• Multi-Model Pipelines (MMP): Enables chaining multiple AI models (e.g., ASR → LLM → TTS) into a single inference pipeline.
• Hardware Acceleration: Leverages Metal, Apple Neural Engine (ANE), and CUDA for optimized performance.

Maintenance & Community

The project maintains an active community presence via Discord and X (Twitter). Contributions are welcomed, with guidelines provided in CONTRIBUTING.md, and specific tasks are tagged for community involvement.

Licensing & Compatibility

Xybrid is released under the permissive Apache License 2.0, making it suitable for commercial use and integration into closed-source projects without copyleft restrictions.

Limitations & Caveats

Support for Bring Your Own Model (BYOM) is experimental. Several features and SDKs are marked as "Coming Soon" or "Planned," including Swift SDK MMP support, Unity MMP support, and specific model integrations like Phi-4 Mini and various embedding models. MMP support is not yet available for Kotlin, Swift, or Unity.