Ollama v0.19

Ollama v0.19 Introduction

Ollama v0.19 is a significant update that rebuilds local AI inference on Apple Silicon using Apple's MLX framework, delivering substantial speed improvements for coding and agent workflows. It introduces NVFP4 support for higher quality responses and smarter cache management, including snapshots and eviction, to enhance performance and reduce memory usage.

Key Features

Massive speedup on Apple Silicon devices through MLX integration

NVFP4 support for improved model quality and production parity

Smarter cache reuse, snapshots, and eviction for efficient memory management

Enhanced KV cache hit rates for faster API responses

Web search plugin for real-time information integration

Use Cases

Developers using local AI models for coding assistance and automation tasks

Data scientists performing real-time analysis with integrated web search capabilities

QA testers ensuring reliability in AI-driven tool calls and debugging workflows

Agile teams accelerating development cycles with faster model iterations and responsive sessions

Why Startups Use It

Startups can leverage Ollama v0.19 to reduce reliance on expensive cloud AI services, lowering operational costs and enabling faster prototyping. The local performance enhancements allow for quicker iterations and more responsive AI applications, which are critical for agile development and competitive advantage.

Frequently Asked Questions

What devices are compatible with Ollama v0.19?

It is optimized for Apple Silicon Macs with unified memory, such as M5 chips, and requires devices with over 32GB of memory for best performance.

How does the MLX integration improve performance?

By leveraging Apple's unified memory architecture and GPU Neural Accelerators, it reduces memory utilization and accelerates both time to first token and generation speed.

Can I use custom models with Ollama v0.19?

Yes, the update supports future models and plans to simplify importing custom fine-tuned models for supported architectures.