Reflection

Reflection in the sidebar opens your Snapshot pages (some screens still call them Continuous Insights). The first page, Snapshot, shows up to 32 cards, arranged two per row. Each card is a prompt, a question you’d otherwise ask every morning, answered fresh from your own sources and the web. You start with three:

Last 30 days: a summary of your conversations across mail, Slack, meetings and chats: the main threads, who they were with, what was decided and what is still open.
Top to-dos: what you committed to or were asked to do and haven’t done yet, most urgent first, with due dates when there are any.
Pending follow-ups: questions to you still unanswered, replies you promised, and things you asked others for that haven’t come back, most overdue first.

Rename the page

Click the pencil beside the Snapshot heading, enter a page name, and click Save (or press Enter). The heading and the page’s tab use your saved name next time you open Olyvr. Choose Cancel or press Escape to keep the current name.

Add more pages

Click New page at the end of the tabs under the page heading, enter a name, and click Create page. The new page uses the same two-column layout and starts empty; add cards just as you do on Snapshot. Each page has its own name, prompts, and answers, saved between app sessions. Switch pages using the tabs. Continuous Insights in the sidebar takes you back to the page you had open last.

You can have up to 8 pages total, including the original Snapshot, with up to 32 cards per page. New page becomes unavailable when you reach the page limit.

To delete a page you added, open it and click the trash button beside its name, then confirm. Its cards are deleted with it. The original Snapshot page can’t be deleted.

Cards

Add a card: click Add card at the top of the page or Add a card after your last card, give it a title and a prompt, and save. It is answered right away. A card can draw on your data, the web, or both, for example “Who is waiting on a reply from me?”, “How the S&P 500 and the dollar did today” or “News about the companies in my open deals”.
Improve: once a card has an answer, hover it and click Improve. The model reads the prompt together with the answer it just produced and rewrites the prompt to fix what the answer missed, drifted into or made hard to read at a glance — without writing that answer’s facts into it, since the data changes each run. It reads your Awareness first — your notes, learned preferences and the preferences that apply to the card — so the rewrite fits who you are, your projects and how you like things shown. The rewrite comes back as a short intent line and a bulleted list of concrete instructions (“show the top 10 most recent changes”, “show items at risk”, “prefer the Jira feed”), one per line, so you can add or remove a line without rewriting a sentence. It opens in the editor: Save runs it, Undo puts back the old prompt.
Edit: change a card’s title, icon or prompt. The prompt only shows while you edit. A new prompt clears the old answer and asks again.
Refresh: asks again now. While Olyvr is open, every card on every page is also answered again every 30 minutes; change how often, or turn it off, in Settings → General → Auto-refresh Snapshot cards. A page with many cards no longer stalls when one model is throttled: refreshes run through a shared queue (up to 3 at a time by default, raise or lower in Settings → General → Snapshot cards refreshed at once, range 1–5), and when more than one model is enabled Olyvr rotates through them, sitting a model out for a minute if it rate-limits so the next card uses the next model. While a card is being refreshed its last answer stays visible with a small spinner, so you can read it instead of a blank card. Refresh all at the top of the page asks every card on it again. A card that hasn’t been answered yet shows Run now.
History: hover a card and click History to see what it asked and showed before. Each time a card is refreshed, its prompt is edited, an Improve rewrite is saved, or the card is deleted, the version that is being replaced is kept first: the prompt, the answer and when it was answered. Choose Use these instructions on an earlier version to put that prompt back in the editor, Forget to drop one version, or Forget all for the card. A card keeps its last 30 versions; across every page, the newest 2,000 stay. Improve also reads the card’s earlier prompts and answers, so it can borrow what worked and avoid what already drifted. In a chat, the assistant can look up what a card showed before (“what did my news card say yesterday?”), and treats it as history, not as current fact. Turn it off in Settings → General → Keep Snapshot card history: nothing new is kept and the assistant can no longer look it up; what was already kept stays until you forget it.
Delete: removes the card, on Snapshot or any page you added. Hover the card and click Delete (or Delete while editing it), then confirm. Its last version stays in its slot’s history.

Arrange cards

Click Edit layout at the top of the page. The cards show a grip, a × to remove them and their size, and the rest of the card controls step aside until you’re done.

Move: drag a card onto another card to take its place; the cards between shift over. Drag it onto Add a card to put it last.
Resize: pick a size on the card (1×1, 2×1 wide, 1×2 tall or 2×2 large), or drag its bottom-right corner; the size snaps to whole columns and rows.
Page layout: the buttons in the blue bar set how many columns and rows of cards fit on a screen.

Click Done (or press Escape) to go back to reading. Changes are saved as you make them.

Cards are answered by the same engine and model as your chats. Like a chat, a card has your Knowledge notes and preferences as context, and it can search and read your workspace’s sources and past chats, and use plugins you’ve marked read-only. It also has a few live sources:

Weather: Open-Meteo.
News: Google News.
Web search: DuckDuckGo.
Any public web page the model chooses to read.

Sources a card cites from your workspace open like documents in a chat. A card can only read: it never creates documents, proposes edits, or runs a plugin action that would need your approval.