MANUAL
The assistant (optional)
Turn it on in Assistant → Assistant Settings… and choose what it runs on:
| Provider | What you need |
|---|---|
| Claude Code | Claude Code installed and logged in (your Claude plan) |
| Codex | The Codex CLI logged in with ChatGPT |
| Anthropic API | An API key |
| OpenAI-compatible | An API key or a local server: OpenAI, Gemini, OpenRouter, Cerebras, Ollama, llama.cpp… (no web search) |
What it can do:
- Research a ⚑ mark: in the In the text pane, press Research. It searches the web, answers with the fact and brings back candidate sources, each with the exact quote that proves it. Cite here accepts one and places it by the mark; Save to Sources just keeps it.
- Critique: Ctrl+Shift+C for the section you're in; Critique the Whole Text is in the Assistant menu. You get 3 to 7 remarks, shown as ✦ in the text and in the pane. What it looks for depends on the template:
- an essay: thesis, logic, evidence and the missing counterargument (in Toulmin: claim, grounds, warrant, limits and rebuttal);
- a blog post: the hook, the structure and the call to action;
- a script or a speech: whether it works by ear, the pacing, and the opening or the ending.
- Questions instead of critique: in an Exploratory essay and in Free writing, the assistant doesn't correct. It leaves 3 to 7 open questions to take your thinking further.
- Versions of a passage, inside the Versions window (see § 10). For scripts and speeches it proposes lines that are easy to say out loud.
- Chat about your text (Ctrl+Shift+A): it sees the current text, the outline and your notes, and knows what kind of text it is. If you select a passage first, the chat is about that passage. Any answer can be inserted where you were writing, or sent to your Notes (→ Notes).
While it works, a window shows what it's doing (what it searches, which page it reads), on how much text, with which provider, and for how many seconds. It has Stop. Critique and research also have Keep writing: the task goes on in the bottom bar and tells you when it's done. The chat shows its progress inside its own pane.
The assistant never writes files or changes your text by itself. It only receives what you ask it to work on, and only the service you chose receives it. Your API keys are encrypted by your system's keychain. Details in SECURITY.md.