Writing Assistant

The Writing Assistant is the AI panel that sits beside your editor. It is best for the moment when you are already inside a note and want help with the thing in front of you: tightening a paragraph, turning rough notes into a draft, finding a clearer structure, or continuing an idea without leaving the page.

Before using it, make sure at least one AI provider or local model is configured. See Configuring AI for setup.

Opening The Assistant

Press Cmd/Ctrl + J to open the Writing Assistant. Octarine uses your current note as the starting context, so you can ask natural things like:

• "Summarize this into a short intro."
• "Rewrite the selected section in a calmer tone."
• "Turn these bullets into a draft."
• "What is missing from this argument?"

If you select text before opening the assistant, only that selection is sent as the main context. You can also select text and choose Add to chat from the Bubble Menu.

Adding More Context

Use @ in the prompt box to add extra context from your workspace. You can mention:

• Notes when you want the assistant to follow a specific reference.
• Folders when a whole project, topic, or archive should shape the response.
• Skills when you want reusable instructions for style, structure, or review behavior.
• Multiple sources when you want it to compare, combine, or write in the same style as existing notes.

The mentions appear as small chips in the prompt box. They stay attached when you edit a prompt, which is useful when the question was right but the context needed one more note.

See Skills if you want to create reusable instructions for prompts you repeat often.

You can remove the current note from context too. That helps when you want a general answer, a fresh idea, or a rewrite based only on the notes you mentioned.

Working With Long Conversations

Each message remembers the prompt, the model, the time, and the context used for that turn. If you switch notes or change the context while a chat is already underway, Octarine marks that the context changed and keeps the conversation grounded in the current material.

You can click a previous prompt to edit it. The prompt opens with its original text and @ mentions intact, then the assistant generates a new answer from the revised request.

Research Mode

Research Mode is for prompts where a quick rewrite is not enough. Instead of rushing into a final answer, the assistant can ask clarifying questions, gather what it needs, and then offer useful next steps such as drafting, summarizing, expanding, or continuing the research.

It is a good fit for:

• Planning an essay, article, or release note from loose source material.
• Turning several notes into a more coherent outline.
• Asking for feedback before committing to a structure.
• Exploring a topic when you are not sure what the final shape should be yet.

Web Search

Web search lets the assistant include current web sources in its response. Click the globe icon below the prompt box to turn it on or off.

Web search is available for OpenAI, OpenAI-compatible providers that support it, Claude, and Codex-backed models. When it is not available for the selected model, the toggle is disabled.

Using Responses

After a response comes back, you can:

• Copy it to the clipboard.
• Insert it at your cursor.
• Replace the selected text with the response.
• Retry to get another version.
• Delete it from the chat.

For writing work, Insert is usually the gentlest option. It lets you compare the suggestion against your original text before deciding what stays.

Slash Commands

Create shortcuts for prompts you reuse often in Settings -> AI Assistant -> Slash Commands. Octarine stores them in .octarine/ai/slash-commands.json inside your workspace.

Slash commands are good for reusable prompt text. Skills are better when you want reusable behavior that can be combined with different prompts and different notes.

Good slash commands are specific and repeatable:

• /tighten for making selected prose shorter without changing meaning.
• /outline for turning messy notes into headings and bullets.
• /release-note for converting a technical note into user-facing copy.

For most writing tasks, a lightweight model is enough. Save the heavier models for deeper research, larger context, or prompts where the assistant needs to reason across several notes.