Install Ollama on your Mac and download a model, then in LocalInbox open Settings › Intelligence, set Written by to “A model server on this Mac” and click Ollama. LocalInbox fills in http://127.0.0.1:11434/v1. Choose your model and click Test Connection. Summaries, replies, drafts and answers then come from that model, on your Mac.
Ollama for macOS, from ollama.com, and enough memory for the model you choose. Apple Intelligence isn't needed.
Steps
- Download Ollama from ollama.com, move it to Applications and open it. It runs in the menu bar.
- Download a model. In Terminal, for example: ollama pull llama3.2
- In LocalInbox, open Settings › Intelligence and set Written by to “A model server on this Mac”.
- Click Ollama. The address becomes http://127.0.0.1:11434/v1.
- Choose your model in the Model menu, or leave Server's first.
- Click Test Connection. It shows which model answered and how long it took.
Server settings
Server address: http://127.0.0.1:11434/v1, Ollama's OpenAI-compatible address on your Mac. If you changed Ollama's port, use yours, ending in /v1.
Troubleshooting
Test Connection says nothing is answering
Ollama isn't running. Open it from Applications (or run ollama serve in Terminal), then try again.
The Model menu is empty
Ollama has no models yet. Download one with ollama pull, then click the reload button next to the menu.
Summaries are slow
The first request loads the model into memory, so it takes longest. If answers stay slow, try a smaller model.
Questions
Does my mail leave my Mac?
Not through LocalInbox: it accepts only a server address on this Mac (127.0.0.1, ::1 or localhost), skips any system proxy and follows no redirects. Choose a model that runs on your Mac, though. Some servers can also hand requests to models in the cloud, such as Ollama's cloud models, and LocalInbox can't see that happen.
What does the model write?
Summaries, suggested replies, Draft with AI, Rewrite and Ask your mailbox. Turn summaries and suggested replies off separately in Settings › Intelligence; what you ask for is still written. Sorting mail doesn't use the server.
Which model should I download?
Any chat model Ollama can run. Smaller models answer faster and need less memory; larger ones write better. Test Connection shows how long yours takes.