Ask Octarine

Ask Octarine lets you chat with your notes to receive summaries, insights, and answers to your questions. It uses a locally-stored embedding model to analyze your workspace content, providing intelligent responses while maintaining complete privacy.

Getting Started

Before using Ask Octarine for the first time, a small 90MB embedding model is downloaded to create vector representations of your notes. This runs entirely on your device — no data is sent to any server for embeddings.

• Access: Click the Ask Octarine sidebar button, press Cmd/Ctrl + O, or use Cmd/Ctrl + K → Ask Octarine
• Initial setup: Octarine downloads a 90MB model to run embeddings on-device. This is fast, cost-efficient, and privacy-friendly since it doesn't use AI credits. If the download doesn't complete, press Manual Setup to download the model files manually and place them in the models directory.
• Indexing: All workspace notes are processed into searchable text chunks on first use. After that, only new, renamed, moved, or deleted notes trigger re-indexing to keep the database in sync.

Adding Context with @-Mentions

Use the @ key in the prompt box to narrow your query to specific notes, folders, or time periods:

• Notes & Folders: Mention specific notes or entire folders to focus the AI on particular content
• Created Date / Modified Date: Filter context by when notes were created or last modified
• Daily Desk Range: Select a date range to include Daily Desk entries from that period

Selected mentions appear as inline chips in the prompt box. The dropdown organizes options into categories — when no search text is entered, you'll see submenu sections for date filters, date ranges, and files & folders. Already-selected items appear at the top with a checkmark.

Octarine is language agnostic — ask in your language of choice and receive answers back in the same.

Chat Interface

Each chat session supports multiple messages with different AI models or contexts:

• Context indicators: Each message displays the context used — folders, notes, date filters, or date ranges — with icons and hover cards showing full details
• References: Listed references show which notes were considered for your query
• Export options: Copy responses as Markdown (for Octarine notes) or plain text
• Chat titles: Automatically generated after the first response, editable via the Sparkles icon
• Save the entire chat to a note by clicking the Create note icon on the chat breadcrumb — user queries are saved as blockquotes, with each message pair separated by a line divider

Editing Previous Prompts

Click on any previous prompt to edit it. The editor re-opens with your original text and any @-mention chips intact. Press Escape to cancel the edit. The AI generates a new response based on your updated prompt.

Prompt Box Layout

The prompt box is split into two sections:

• Top section — The text editor where you type your prompt and add @-mentions
• Bottom toolbar — Contains the model selector, web search toggle, and help button, with the send/stop button in the editor area

Suggested Prompts and History

Empty chats display suggested prompts based on a random selection of folders with notes. Click Refresh Suggestions to get new options.

History entries are organized by relative dates (Today, Yesterday, 2d ago, 1w ago) and maintain their references when accessed from history.