MailTwin
Which AI should I pick?
MailTwin doesn't lock you into a model — you bring your own key, and you switch any time from the panel footer. But staring at eight options when you just want to reply to an email is its own kind of friction. Here's the short version.
If you want zero setup: Apple Intelligence
Built into macOS 15.1+ on any M-series Mac. No key, no bill, no data leaves your Mac. Best for short replies, summaries, action items, and tone changes — the model is small (around 3 billion parameters) so it's noticeably less capable than the cloud models on long-form drafting in your voice. We pre-select this for you if your Mac supports it.
If you care about quality: Claude or Gemini Pro
These two are the gold standard for writing-style matching right now. Claude (Sonnet 4.6 or Opus 4.7) reads your style profile and produces replies that genuinely sound like you wrote them yourself. Gemini 2.5 Pro is right behind, with a generous free tier — sign up at aistudio.google.com and pay nothing for typical use.
If you want completely local: Ollama
Run any open-weights model on your Mac (Llama 3.3 70B, Qwen,
DeepSeek-V3, whatever you've pulled). No data leaves the machine,
nothing routes through us. Slower than cloud providers and the
smaller models can be hit-or-miss, but it's the privacy
gold-standard. Install Ollama from
ollama.com, run
ollama serve, point MailTwin at it.
The full menu
| Provider | Cost | Best at | Get a key |
|---|---|---|---|
| Apple Intelligence | Free, on-device | Quick replies, summaries | (none — built in) |
| Ollama | Free, on-device | Privacy-first, any open model | (none — local server) |
| Groq | Free tier, your key | Fastest replies (Llama 3.3 70B) | console.groq.com |
| Google (Gemini) | Free tier, your key | Long-form, multilingual | aistudio.google.com |
| Anthropic (Claude) | Paid, your key | Writing in your voice | console.anthropic.com |
| OpenAI (ChatGPT) | Paid, your key | General-purpose, broadly capable | platform.openai.com |
| xAI (Grok) | Paid, your key | Casual, less filtered tone | console.x.ai |
| DeepSeek | Paid, your key | Cost-efficient cloud reasoning | platform.deepseek.com |
Switch any time
The provider picker lives in the panel footer next to your trial / license badge. Click it, pick a different provider, MailTwin uses that one for the next reply. No restart, no rebill, no re-onboarding. Different providers for different accounts works too — set it once per account in Settings → AI → Providers.
What happens when a model gets retired
AI providers deprecate models on short notice. MailTwin checks a remote config file every four hours; if your active provider flagged a model you're using as deprecated, the app surfaces a clear "switch provider" banner with a one-click migration to a working option. No app update needed. No silent failures.