With Ollama as the provider, dossier’s AI runs entirely on your Mac: no API key, no per-use cost, and no document text ever leaves your machine.
Install and pull a model
brew install ollama
ollama pull llama3.3
Any chat-capable model works; bigger models give noticeably better filing proposals.
On an Apple-silicon Mac with 32 GB+ RAM, a 70B-class model like llama3.3 is a good
default. On smaller machines try llama3.1:8b or qwen2.5:14b — quicker, but expect
to correct more proposals.
Start the server
OLLAMA_ORIGINS="*" ollama serve
(The OLLAMA_ORIGINS part avoids CORS errors when dossier talks to the local server.
If you run Ollama.app instead of the CLI, it usually just works.)
Connect it
- In dossier: Settings → AI → Provider: Ollama (local).
- URL:
http://localhost:11434(the default). - Pick from the models you’ve pulled — dossier lists what’s installed.
What to expect
Local models are great for proposals and searching, and the assistant works with them too — dossier includes extra tolerance for the ways smaller models fumble tool calls. For the smoothest assistant experience, a hosted model is still ahead; many people mix: local for everyday filing, hosted for the occasional heavy question.