Ask Octarine
Chat with your workspace, search across notes, and turn scattered context into useful answers.
Ask Octarine is the workspace-level AI chat. Use it when the answer is probably somewhere in your notes, but you do not want to manually search, open five files, and stitch the story together yourself.
Ask Octarine is only available for users on the Pro License.
What It Does
Ask Octarine can answer questions, summarize projects, find old decisions, compare notes, and help you create new writing from what already exists in your workspace.
With a tool-capable model, Ask Octarine now works more like a small research agent. It can search your notes, read the most relevant ones, follow backlinks and wikilinks, inspect tags, list folders, and then answer from what it found. This is different from a simple one-shot search: it can take multiple steps when the question needs it.
For example:
- "What did I decide about pricing in the last two months?"
- "Summarize everything related to the mobile redesign."
- "Find notes that mention the export bug and tell me the current status."
- "Create a brief from @Projects/Website and @Meetings/June."
- "What open threads did I leave in Daily Desk last week?"
If the selected model does not support tools, Octarine falls back to a simpler single-shot search. You will still get answers from your notes, but the full agent path is better for broad, messy, or multi-hop questions.
First Setup
The first time you use Ask Octarine, Octarine downloads a local embedding model. Embeddings make your notes searchable by meaning, and they run on your device.
- MiniLM L6 v2 is the default model. It is smaller, faster, and about 90MB.
- Nomic Embed v1.5 is the higher-quality option. It is about 140MB and takes a little more time and resources to index.
- Your notes are indexed locally.
- New, renamed, moved, edited, or deleted notes are re-indexed automatically.
- Only your actual chat query and the relevant note content are sent to your selected AI provider.
To keep the whole flow local, use a local provider such as Ollama or LM Studio.
Opening Ask Octarine
Open it from the sidebar, press Cmd/Ctrl + O, or use Cmd/Ctrl + K -> Ask Octarine.
If indexing is still running, the prompt box shows progress. You can start asking once the workspace is ready.
Excluding Notes And Folders
Use Ask Octarine exclusions when some workspace content should stay out of AI search. Excluded notes and folders are not indexed by Ask Octarine and will not appear as @ context suggestions.
Open Settings -> AI -> Ignored, then add the folders or notes you want Ask Octarine to skip. Removing an item from the ignored list makes it available again.
When anything is ignored, the prompt box shows an N ignored shortcut. Click it to jump back to the exclusions settings.
Exclusions are saved in a .octarineignore file at the root of the workspace, with one path per line. You can manage the list from settings, or edit the file directly if you prefer.
For example:
# Ask Octarine exclusions
Private
Research/Unpublished.md
.templates
Ignoring a folder excludes everything inside it. This applies to workspace search, similar note lookup, backlinks, wikilink expansion, note outlines, and direct note reads inside Ask Octarine.
Controlling Context
Ask Octarine can search the whole workspace, but you can narrow it when you already know where the answer should come from.
Type @ in the prompt box to add:
- Notes for exact source material.
- Folders for projects, areas, or archives.
- Skills for reusable instructions, formats, or review lenses.
- Created or modified date filters for time-bound questions.
- Daily Desk ranges for questions about a week, month, trip, sprint, or other stretch of days.
You can combine several contexts in one prompt. For instance, ask it to compare one folder with another, or use a Daily Desk range plus a project folder so the answer is both time-aware and topic-aware.
Skills behave a little differently from notes and folders: they guide how Ask Octarine should work, while the other mentions tell it what material to use. See Skills for setup and examples.
Octarine is language agnostic. Ask in the language you prefer, and it will answer in the same language.
Web Search
If the selected model supports it, you can turn on web search from the globe icon in the prompt box.
Use web search when your notes are the starting point but not the whole answer: checking current facts, filling in public context around a saved link, or comparing your own notes with something that has changed since you wrote them.
When web search is unavailable for the selected model, the toggle is disabled.
Reading The Answer
Agent answers may include the notes it used as clickable wikilinks. The chat also keeps a reference list so you can inspect the source material yourself.
Each message keeps its own context, model, and references. That means one chat can include a broad workspace question, a follow-up narrowed to a folder, and then a third question using a different model without losing track of what happened.
You can copy responses as Markdown or plain text. You can also save the whole chat to a note from the chat breadcrumb; Octarine writes your prompts as blockquotes and separates each exchange with a divider.
Editing Prompts
Click any previous prompt to edit it. Octarine reopens the text with the same @ mentions, then regenerates the answer from the updated request.
This is especially useful when the first answer was close but you want to add a missing folder, tighten the date range, or ask for a different format.
Suggested Prompts And History
New chats show suggested prompts based on a random selection of folders with notes. Refresh them when you want a different starting point.
History is grouped by relative date, such as Today, Yesterday, 2d ago, and 1w ago. Saved chats keep their references, so you can return to an answer and still see what it was based on.
Picking A Model
For the full Ask Octarine agent, choose a model that supports tool calling. Octarine will warn you when the current model can only use the simpler search path.
Use lighter models for quick summaries and straightforward lookup. Use stronger models when the question spans many notes, asks for synthesis, or needs careful reasoning across projects.