Premail

AI Model Configuration

Last updated

Premail classifies your email using a large language model. It sends the sender, subject line, and message body to the model and asks it to categorize the message. You pick the provider.

The Premail AI Model screen with Anthropic, OpenAI, Ollama, and Apple Intelligence options. OpenAI is selected, showing a masked API key field, a model picker set to gpt-4o, a Refresh button, and running cost estimates for the current model and all time.
Pick your provider and paste your own key, or run Ollama locally so nothing leaves your machine. Premail tracks what the calls cost you.

How do you use Anthropic (Claude)?

The default. Premail uses Claude Haiku, which is fast and accurate. It costs roughly $11 per 1,000 emails classified, so a busy inbox adds up. Get an access key from console.anthropic.com and paste it into the AI Model tab.

The Premail AI Model screen showing OpenAI / gpt-5-mini in use with running costs of $1.05 for the current model and $58.40 all-time, the Anthropic, OpenAI, Ollama, and Apple Intelligence provider row, a green "Apple Intelligence is ready" panel, and a Model calls per hour field set to 500.
Premail tracks what your own API key is actually costing you, per model and all time, and caps how many paid calls it will make in an hour so a busy inbox cannot run up a surprise bill.

How do you use OpenAI?

Works the same way. Paste in your OpenAI access key, pick a model. GPT-5 Mini is the default and a reasonable choice if you're already in the OpenAI ecosystem and don't feel like creating another account somewhere.

How do you use Ollama locally?

Ollama runs AI models locally. Nothing gets sent to any external server, which may matter to you a great deal or not at all, depending on your personal threat model.

  1. 1Download and install Ollama from ollama.com. A small icon will appear in your menu bar after installation.
  2. 2Open Terminal and pull a model. We recommend ollama pull qwen3 for the best classification quality on local hardware.
  3. 3In Premail's AI Model tab, select Ollama and enter the model name you pulled (e.g.qwen3).

Performance depends on your hardware. Classification will be slower than cloud providers, and quality varies by model. Larger models are more accurate but slower. qwen3 strikes the best balance we've tested for Premail's filtering tasks.

How do you use Apple Intelligence?

On a Mac running macOS 26 or later with Apple Intelligence support, Premail can classify mail using the on-device Apple Intelligence model instead of a cloud provider or Ollama. There is no API key, no separate AI account, no model to download or choose yourself, and no Ollama install - macOS manages the model for you.

  1. 1Turn on Apple Intelligence in System Settings, if you haven't already. The pane is called Apple Intelligence & Siri on macOS 26 and just Siri on macOS 27 and later (where Apple Intelligence is on whenever Siri is on).
  2. 2Let your Mac finish obtaining the on-device model assets. This happens in the background and depends on your network connection.
  3. 3Return to Premail, select Apple Intelligence in the AI Model tab, and click Check again. Premail reads live readiness from macOS rather than a saved setting, so the panel updates once your Mac reports it is ready.
  4. 4Confirm it works by dropping a sample .eml file onto Dropzone and checking the classification result.
See the full Apple Intelligence setup guide for requirements, what "on-device" does and does not cover, and troubleshooting.
Your access keys are stored locally on your computer and are never sent to any Premail server. They go directly from your machine to whichever AI provider you've configured. Apple Intelligence needs no key at all.
←Start at LoginApple Intelligence Setup→

Still have questions?

Contact Support (for Pro members)