OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Local & Self-hosted AI

TurboFieldfare

TurboFieldfare is an open-source Swift and Metal runtime for running Gemma 4 26B-A4B locally on Apple Silicon Macs. It provides a native Mac app, a command-line interface, an experimental loopback OpenAI-compatible server, and a Swift library. Its memory design keeps a shared model core and cache in memory while streaming selected experts from SSD.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

TurboFieldfare — OpenViST project profile in Local & Self-hosted AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use TurboFieldfare when you want to run its supported Gemma 4 model locally on an Apple Silicon Mac and have the required macOS, Metal, development tools and storage.

Who it is for

  • Apple Silicon Mac users who want to generate text with Gemma locally.
  • Developers seeking a local Gemma inference runtime, CLI or experimental server for Mac applications.

Documented uses

  • Install Gemma 4 26B-A4B and generate text through a native Mac app.
  • Use instruction chat or raw completion from the command line.
  • Expose local model generation through an experimental loopback Chat Completions server.
  • Use the Swift runtime library as a basis for Mac-side local inference work.
Before you choose

Requirements & limits

The repository is public and licensed under Apache-2.0. The documented build requires an Apple Silicon Mac, macOS 26, Metal 4, Xcode 26 and Swift 6.2 or newer.

  • The runtime targets Gemma 4 26B-A4B and is model-specific rather than a wrapper around MLX or llama.cpp.
  • The roughly 2 GB figure describes the stated weights-and-cache memory budget, not the disk footprint; the text model installation is about 14.3 GB.
  • The server is described as experimental, and its protocol and integration details are not specified here.
  • The server can return model-produced tool calls, but the client must authorize and execute them; the app and CLI do not expose or execute tools.
  • Image input needs a separate companion vision pack; audio and video are unsupported.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Build the Mac app and prepare the Gemma model

Build TurboFieldfare, install its model, and generate text in the native Mac app.

Before you start

  • Apple Silicon Mac
  • macOS 26 with Metal 4
  • Xcode 26 and Swift 6.2 or newer
  • About 14.3 GB free storage for the text model
  • Internet connection for the first model installation

Steps

  1. Clone the TurboFieldfare repository and change into its directory.

  2. Build the release app with swift build -c release.

  3. Launch .build/release/TurboFieldfareMac; on first launch, Swift Package Manager downloads and builds required tokenizer packages.

  4. Choose Download and let the app fetch and repack the model, then choose Load Model, enter a prompt and select Generate.

Expected outcome

The native Mac app generates text using the locally installed Gemma 4 26B-A4B model.

Guide limitations

  • The text model installation is about 14.3 GB; image input requires a separate optional image pack.
  • Older macOS and Metal versions are unsupported, and the package is arm64-only.
  • Generation can be incorrect or repetitive; check important results.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
7 October 2026
Original website title
drumih/turbo-fieldfare · Official homepage title · retained in its source language
How we review sourcesSuggest a correction