Outmake has no account of its own. Each agent you chat with is your account with that company, and Settings › Accounts is where you connect them.
Accounts and keys
Settings › Accounts lists Claude (Anthropic), ChatGPT (OpenAI, through Codex), Gemini (Google) and OpenCode, plus This Mac, which needs no account at all. Signing in never happens inside Outmake: Claude signs in in a Terminal window Outmake opens, running Anthropic’s own Claude Code, and ChatGPT on OpenAI’s page. Gemini takes an API key. An API key you paste in is kept in your Mac’s keychain and given only to the agent it belongs to.
Claude and ChatGPT run on Node.js. If your Mac doesn’t have it, Outmake downloads its own copy from nodejs.org with your first message (about 53 MB, checked before it runs) and keeps it in its own folder; a Node.js you install yourself is always used first. Settings › Accounts can download it ahead of time. OpenCode and models on This Mac don’t need it.
OpenCode is its own agent: point it at any API key or a model server you already run, including one on this Mac served by Ollama or LM Studio. Claude and ChatGPT don't reach those servers themselves.
Settings › Models › Keys keeps keys for other model companies: Add a Provider lists DeepSeek, Moonshot, Kimi, Z.ai, xAI, MiniMax, Xiaomi MiMo, Qwen, Meta, Vercel AI Gateway, OpenCode Zen and more. Their models then appear in OpenCode chats. Where the company says, the key’s row shows how much credit is left. Plans signs in to a subscription such as GitHub Copilot through OpenCode’s own sign-in. Commands an agent runs never see these keys.
What is sent where
Your messages, attachments and screenshots in a chat go to the company behind that chat's agent, under your account with them. A chat on This Mac is the one exception: nothing leaves this Mac at all. Each row under Settings › Accounts also has its own Uses your memory switch, which is what lets that agent see your memory (see Memory).
The consent question
Outmake asks before anything goes to another company for the first time: a dialog naming the company and what it would get, such as "Send your chats to Anthropic?", with Allow or Don’t Allow. It asks once per company and per use: saying yes for chats doesn’t also say yes for pictures or speech. Your answers live in Settings › Privacy, one switch each, and turning one off stops the very next send. See Privacy.