Why Kit needs a model
What Kit does with a model
- Talking. Your chats with Kit, in the app and on Telegram, and sorting the mail you receive.
- Filing. Turning your conversations, mail and files into memories, and writing Kit's summaries of the people, places and topics in them.
- Dreaming. Kit's nightly memory work: merging what repeats, linking what belongs together, and retiring what is out of date, so the morning starts tidier than the night before.
- Checking. Second opinions on what filing wrote, Kit's own notes, and the nightly checks that measure how well it still recalls.
What needs no model: storing your memories, searching them, and recognising names. Those run on your Mac, whichever choice you make below.
Three ways to run it
Kit's model, the default
Nothing to set up. Kit's model is a gateway we run, hosted by Hetzner in Ashburn, Virginia, in the United States, that passes each job to OpenAI or Anthropic and brings the answer back. Today, Anthropic's Claude does the talking and checking, and OpenAI's GPT does the filing and most of the dreaming. The exact models, and the backups used when one is unavailable, are listed in the terms, and that list changes there first.
That means the words a job needs, your memories and messages included, are sent to OpenAI or Anthropic. The gateway keeps no prompts or answers, only a usage record for 90 days so it can apply the monthly allowance. Neither provider trains on it, and both normally delete it within 30 days. During the beta, each Mac gets $5 of Kit's model a month, an introductory offer that may change or end.
Your own key
Bring your own Anthropic or OpenAI key, and Kit's prompts go straight from your Mac to that provider, under your own account and their terms. Nothing passes through our servers, and the cost is your provider's bill.
A model on your Mac
Run a model through Ollama or LM Studio and nothing leaves your machine. Switch on Local only, and Kit refuses any call that would. The trade is quality and speed: a local model is slower, and a frontier model writes better memories.
Mix them, and change your mind
You don't have to pick one for everything. In Settings, AI models, each job (talking, filing, checking) can use a different model, and the Models pane shows which calls leave your Mac and what they cost. You can switch at any time, and your memory stays where it is.