Writing with Claude Code
Flywrite doesn't bundle its own AI. Instead, you run the real Claude Code (or Codex, or any CLI agent) in the built-in terminal, on your own subscription — with your skills, your memory, and no extra AI fees from Flywrite.
Before you start
Install the agent you want to use — for Claude Code, follow Anthropic's install guide. Once it runs in your computer's terminal, it runs in Flywrite.
The writing loop
- Open a terminal and launch Claude Code —
⌃2, or the + button in the terminal tab bar. - Select some text in your document. Claude Code automatically sees your selection and your open files — no copying and pasting.
- Ask for what you want — "tighten this paragraph," "draft an intro from these notes."
- Watch the edits land. The agent edits your files directly, and the editor updates live.
- Review and keep writing. You can hand-edit right alongside the agent.
Pointing the agent at your work
Most CLI agents, including Claude Code and Codex, let you
@-mention a file right in your prompt (type @
and the file name). In Flywrite you often don't even need that: the app
automatically tells the agent which file you're looking at, which line
you're on, and which text you've highlighted — no pasting content into
your prompt.
To hand off a selection explicitly, press ⌘⇧L —
Send Selection to Terminal — and Flywrite drops a reference
to it into the terminal for you. (These shortcuts are defaults — all of them
are customizable.)
Your safety net
Flywrite automatically saves a checkpoint the moment an agent starts working, so you can always see exactly what changed — and restore any earlier version if you don't like it. See Checkpoints & versions.
Know when it's done
A spinner on the terminal tab shows the agent is busy. When it finishes, Flywrite sends a system notification (and can play a sound — toggle it in Settings → Writing). Click the notification to jump straight back to that terminal.