Choose the AI
Pick the AI that sorts your messages, writes drafts and answers your questions, and optionally a local model for private chats.
ChatPilot needs an AI to sort what comes in, write drafts, summarise and answer your questions. You use your own: your Claude subscription, an AI service with your own key, or a model on your own computer. Open Settings → AI.
The main AI
Choose one under Main AI:
| Choice | What you need |
|---|---|
| Claude Code (your Claude subscription) | Claude Code installed and signed in (claude, then /login). No separate bill. This is the default. |
| Anthropic API (Claude) | an API key from Anthropic |
| OpenAI API (ChatGPT models) | an API key from OpenAI |
| xAI API (Grok) | an API key from xAI |
| Google Gemini API | an API key from Google |
| Mistral API | an API key from Mistral |
| Other OpenAI-compatible service or local model | its address (Server; buttons for Ollama and LM Studio) and a key if it needs one |
| OpenAI Codex CLI (sign in with ChatGPT) | the Codex command-line app, signed in (codex login) |
| Google Gemini CLI (sign in with Google) | the Gemini command-line app, signed in (run gemini once) |
Then:
- Paste the API key if it asks for one. It goes into your computer's secret store, never into a file.
- Pick the Model. Find models lists what the service offers. (Claude Code has no model field: it uses the Claude model set in
config.yaml.) - Press Test connection: ChatPilot makes one tiny call and shows the answer.
- Press Save. The new AI takes over at once, without a restart; messages that were waiting are sorted then.
Signing in without a terminal. The three sign-in choices (Claude Code, Codex CLI, Gemini CLI) have a Sign in here button. It runs the program's own sign-in on the computer ChatPilot runs on, which helps when that is a server you only reach through the panel. Press Open the sign-in page and approve there. Codex shows a code to enter on that page; Claude Code and Gemini show a code after you approve, which you paste into the box under the button. Then press Test connection. The program keeps its own login, as if you had signed in in a terminal.
Can do shows honestly what the AI you picked can do: whether it holds to ChatPilot's answer format, whether it may look things up (read-only) for drafts and discussions, whether it can look at images, and whether it runs on this computer or online.
A second model: local or another service
Under Second model: local or another service you can add a model that runs on your own computer (Ollama or LM Studio) or another service with an OpenAI-compatible API:
- Fill in the Server (or press Ollama / LM Studio) and pick the Model.
- Under Use it for, tick the tasks it takes over (for example sorting, summaries and translation).
- Test, then Save.
This model is also the only AI for chats you set to Only the AI on this computer: nothing of those chats goes to the main AI. If the local model isn't running, ChatPilot waits rather than send them elsewhere.
Local models are free and private, but slower and less sharp than the big online models, especially for drafts.
What goes where
Only chats you set to ChatPilot can read this go to the main AI, under your agreement with its maker. See Choose which chats ChatPilot may read.
Checked against ChatPilot 0.1.0b1. Report a problem with this page