Mac assisted food entry

Connect AI food assistance on Mac

Choose Apple Intelligence, OpenAI, Claude, or a vision-capable Ollama model and review every result before saving.

BeFit Tracker can use a configured AI connection to match a dictated description or food photo with useful recipe candidates. The provider helps with discovery; you choose the source, inspect the nutrition, and decide what enters your User Library.

AI provider choices in BeFit Tracker Settings on Mac — Add the connection that fits your Mac and provider account. You can keep more than one connection configured.

Choose a connection

Apple IntelligenceBuilt into supported Macs · Beta

Photo support when the local system model is available. Opt in from Settings.

Uses the system model on your Mac.

OpenAI APISelect a model with image input

Good starting points include GPT-5, GPT-5 mini, GPT-4.1, and GPT-4.1 mini.

Provider account, pricing, limits, and network access apply.

Claude APISelect a vision-capable model

Use a current Claude Sonnet or Opus model, or a supported Claude 3 vision model.

Provider account, pricing, limits, and network access apply.

OllamaA local vision model is required

Try Gemma 3 4B/12B/27B, Llama 3.2 Vision, or Qwen2.5-VL.

Ollama must be installed and running locally.

Compatible examples verified against provider documentation in August 2026. Provider catalogs change, and availability can differ by account and Ollama version. For a photo, confirm that the provider describes the selected model as accepting image input.

OpenAI or Claude

Open BeFit Tracker Settings and add an AI connection.
Choose OpenAI or Claude, enter the API key from your provider account, and select a model.
Save the connection. BeFit Tracker stores the credential in Keychain.
From the User Library, choose a dictated or photographed food workflow and review the returned recipe candidates.

OpenAI model picker in BeFit Tracker on Mac — The model list comes from the configured provider. For a photo, select one the provider documents as accepting image input.

Apple Intelligence on Mac Beta

Apple Intelligence is optional and must pass an availability check before BeFit Tracker offers it as a connection. The workflow combines Apple image classification with the on-device system model to form a useful food description, then searches recipe sources for candidates.

Recognition is a starting point.

Apple Intelligence can misidentify a dish, omit an ingredient, or return an unhelpful description. This Beta workflow is opt-in. Check the recognized food, recipe source, ingredients, serving, and nutrition before saving.

Apple Intelligence selected as an AI connection in BeFit Tracker on Mac — Apple Intelligence is offered only when the system model is available on this Mac.

Ollama stays local for recognition

Run Ollama and install a model whose Ollama page explicitly shows Image input. The selected model analyzes the picture through the local Ollama service. Recipe-page discovery that follows can still require the internet.

For Gemma 3, use 4B, 12B, or 27B. The 270M and 1B tags are text-only. Other supported starting points include Llama 3.2 Vision and Qwen2.5-VL.

Photo-assisted food entry in BeFit Tracker on Mac — After recognition, compare real recipe candidates and keep the final decision with you.

Before anything is saved

Check the dish name, recipe website, ingredients, nutrition, and serving. A model can be convincing and still be wrong. BeFit Tracker does not calculate reliable nutrition from pixels alone and does not guarantee a provider's output.

Provider references: OpenAI image input, Claude vision, and Ollama vision models.