Foundation Models in macOS 27: How Apple’s Local AI Can Automate Business Processes

With macOS 27 Golden Gate, Apple introduced fm — a built-in command-line tool that provides direct access to Foundation Models from Terminal. Modern Macs now have a local language model that can be used in applications, scripts, and automated workflows without relying on a separate AI service, API keys, or per-request fees.

fm runs on Apple Silicon and can respond to prompts, return structured data in JSON, and launch a local API server that other applications can connect to.

fm respond "Write a short description for this product"

Product descriptions and recommendations for e-commerce

In an e-commerce project, the model can take product specifications and automatically generate descriptions, SEO copy, and alt text for product images.

It can also analyse the catalogue and suggest related products: a cushion that matches a sofa, an armchair from the same collection, or a relevant accessory as an upsell.

Instead of manually working through hundreds of products, the store gets a system that can handle these tasks automatically whenever a new item is added. With fm serve, the model can be exposed as a local API and connected to the store’s CMS, ERP, or backend, while the actual processing takes place on a Mac within the company’s own infrastructure.

Advertising reports

Another useful application is automated marketing reporting. A script can collect data from Google Ads, Analytics, or similar platforms, while the model turns those tables and metrics into a clear report for the person running the business:

“Spending increased by 14%, while the number of leads remained unchanged. The main reason is Campaign X, where the cost per lead rose from €18 to €31 over the past week. Campaign Y, on the other hand, is delivering its best results in the last two months.”

The system can identify anomalies, compare different periods, and prepare different versions of the same report for a marketing specialist and for a business owner. Reports can then be sent automatically on a schedule to a CRM, email inbox, or internal dashboard.

Turning a website enquiry into a structured lead

Suppose a potential client sends the following enquiry:

“We want to rebuild our online store. We currently use Shopify, but we need a more customised solution with a recommendation system, better SEO, and a stronger GEO strategy. The catalogue has around 5,000 products, we would ideally like to launch before December, and the budget is approximately €20,000–30,000.”

For a CRM, this would normally arrive as an unstructured block of text. Using Apple’s model and fm schema, it can automatically be converted into structured data:

{
  "service": "e-commerce",
  "budget": "20000-30000",
  "deadline": "December",
  "platform": "Shopify",
  "catalog_size": 5000
}

The data is written directly into the appropriate CRM fields. The manager immediately sees the requested service, budget, deadline, and key project requirements without having to process the message manually. The system can then assign the right person and create the necessary tasks automatically.

The same structured data can also flow directly into analytics tools. This means the sales funnel and advertising campaigns can be tracked in real time, making it much easier to understand where leads are coming from and use the marketing budget more efficiently.

How this can be useful for your business

Local AI models are not new. What matters here is that Apple is making its own Foundation Model part of macOS itself, optimised specifically for the hardware it runs on, while giving developers an official way to integrate it into applications and automated business processes.

For a small business, this can mean less manual work and more automation inside its own private infrastructure, without having to send company data to a third-party cloud service for processing.

Local models in macOS 27 are another practical tool for making business automation simpler, more private, and less expensive.