Skip to content

Local AI testing

Joe Maloney edited this page Nov 15, 2026 · 9 revisions

Installation of the server used for the model using Mac Studio M5 Max base model

brew install incoai/tap/splash

Installing the model and serving the model to the entire network

For 36GB of memory or less

splash serve --model incoai/Qwen3.8-27B-Splash --max-cache-disk 20GB --host 0.0.0.0 --port 8000 --allowed-host mac-studio.home.local

Fore more than 36G of memory

splash serve --model incoai/Qwen3.8-27B-Splash --host 0.0.0.0 --port 8000 --allowed-host mac-studio.home.local
  • The --max-cache-disk 20GB setting was necessary since 48GB is the actual recommended for the model, 36GB will work but this resolves any warnings that memory might not be available when needed.
Screenshot 2026-10-02 at 11 09 45 PM

Configuring OpenCode to connect

edit ~/.config/opencode/opencode.jsonc

{
  "$schema": "https://opencode.ai/config.json",
  "model": "splash/incoai/Qwen3.8-27B-Splash",
  "provider": {
    "splash": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Splash (mac-studio)",
      "options": {
        "baseURL": "http://mac-studio.home.local:8000/v1"
      },
      "models": {
        "incoai/Qwen3.8-27B-Splash": {
          "name": "Qwen3.8-27B (Splash, mac-studio)",
          "reasoning": true,
          "interleaved": {
            "field": "reasoning_content"
          },
          "limit": {
            "context": 256000,
            "output": 32768
          }
        }
      }
    }
  }
Screenshot-2026-10-02-231139

Special .tmux.conf needed for copy paste to properly work if using tmux

# Use vi-style keybindings in copy mode
setw -g mode-keys vi

# Copy selected text to the macOS clipboard using pbcopy
bind-key -T copy-mode-vi y send-keys -X copy-pipe-and-cancel "pbcopy"
bind-key -T copy-mode-vi MouseDragEnd1Pane send-keys -X copy-pipe-and-cancel "pbcopy"

Clone this wiki locally