Local AI Processing

Apple Intelligence, MLX models, and Ollama — free, private, on-device AI.

VivaDicta offers three ways to run AI processing entirely on your Mac — no API keys, no internet, no data leaving your device. Choose the option that best fits your needs.

Apple Intelligence (Foundation Models)

Apple's built-in on-device AI, powered by Apple Foundation Models. The simplest option — zero setup required.

Requirements

• macOS 26 (Tahoe) or later.
• Apple Silicon Mac (M1 or later).
• Apple Intelligence enabled in System Settings → Apple Intelligence & Siri.

Setup

1. Open VivaDicta settings → AI Provider.
2. Select Apple Intelligence.
3. That's it — no API key or sign-in needed.

Best For

• Simple tasks: grammar correction, punctuation, basic reformatting.
• Users who want zero setup and maximum privacy.
• Quick dictation cleanup without configuring any provider.

Limitations

Apple Foundation Models are less capable than cloud models (Claude Sonnet, GPT-5) or larger local models for complex tasks like summarization, translation, or coding. For those, consider MLX or cloud providers.

Apple MLX (Open-Source Local Models)

Run powerful open-source LLMs directly on your Mac using Apple's MLX framework. One-click model download, optimized for Apple Silicon with Metal GPU acceleration. More capable than Apple Intelligence for complex tasks.

Available Models

VivaDicta organizes models by Mac RAM tier so you can quickly find the best model for your hardware:

Quantization Variants

Most models offer Base, Med, and High quantization:

• Base — smallest download, fastest, slightly lower quality.
• Med — balanced quality and speed.
• High — best quality, requires more RAM.

Getting Started

1. Open VivaDicta settings → AI Provider → Apple MLX.
2. Browse models by your Mac's RAM tier. A Recommended badge highlights the best model for your hardware.
3. Click Download on a model.
4. Once downloaded, select it as your active model.

Best For

• Users who want powerful AI processing without paying for API access.
• Privacy-conscious users — nothing leaves your Mac.
• Offline use — works without internet after model download.
• Complex tasks that Apple Intelligence can't handle well.

Ollama

Run any Ollama-supported model (Llama, Mistral, Gemma, Phi, etc.) locally. Ollama provides access to a wider range of models beyond what's built into VivaDicta.

Setup

1. Install Ollama from ollama.com.
2. Pull a model: ollama pull llama3.2
3. In VivaDicta, go to AI Provider → Ollama.
4. No API key needed — connects locally at http://localhost:11434.

Best For

• Users already familiar with Ollama.
• Access to models not available in VivaDicta's MLX catalog.
• Experimenting with different model architectures.

Comparison

System Requirements

• Apple Intelligence: Apple Silicon + macOS 26+.
• MLX: Apple Silicon + enough RAM for chosen model.
• Ollama: Any Mac (Apple Silicon recommended for speed).