← all releases
Inco AI·Nov 19, 2026·v1.7.0·open source

Splash

Splash is a high-performance, open-source local inference engine specifically optimized for Apple silicon. It achieves superior speed by utilizing specialized Metal kernels, automatic memory planning, and DFlash 2 speculative decoding tailored to each supported model.

Local AIInference EngineApple SiliconLLM
visit website github ↗
overview

Splash is a high-performance, open-source local inference engine specifically optimized for Apple silicon. It achieves superior speed by utilizing specialized Metal kernels, automatic memory planning, and DFlash 2 speculative decoding tailored to each supported model.

key features
  • 01Specialized Metal kernels and DFlash 2 speculative decoding
  • 02Automatic memory planning and batching of concurrent requests
  • 03Full compatibility with OpenAI and Anthropic API standards
  • 04Built-in chat interface and support for coding agents
  • 05Vision, tool calling, and JSON Schema output support