Splash
A local inference engine for Apple silicon that runs coding agents and OpenAI or Anthropic compatible applications on one Mac, combining speculative decoding with Metal kernels and automatic memory planning.
Splash is a local inference engine for Apple silicon, installed with Homebrew, that serves models over an OpenAI Chat Completions and Responses compatible API as well as an Anthropic Messages compatible API, with streaming, tool calls, JSON Schema output, images and inline PDFs. It combines DFlash 2 speculative decoding with specialized Metal kernels and automatic memory planning, reuses cached prefixes, and batches concurrent requests automatically.
It requires an Apple M3 or newer chip and macOS 26.4 or later, with larger 4-bit examples needing 36 to 48 GB of unified memory and smaller GGUF variants available for 24 GB Macs. Supported model families include Qwen3.8 and Qwen3.6 variants with matching draft models, and it can serve coding agents such as opencode, Claude and Codex directly from the command line.