Memory (Second Brain)

Memory (the sidebar also calls it Second Brain) is where Everpaper reflects your work back to you. It reads across every day you've captured, your recorded meetings and the folders you include, links the things that keep recurring, and answers your questions with sources.

Open Memory in the sidebar, or choose Go → Second Brain (⌘4). Memory has six tabs, on ⌥⌘1 to ⌥⌘6: Ask, Episodes, You, Insights, Connections and Export. ⌥⌘← and ⌥⌘→ move to the previous or next tab. Memory opens on Ask.

How ready it is#

The top of the page shows Still learning you, Getting to know you or Ready, and how many active days it has seen. Below that are four counts: Days, Frames processed (screenshots), Topics (the distinct people, files, sites and tickets it has found) and Recurring threads (any of these seen on more than one day). Click the arrow beside the counts to see the percentage and a short note. On the Ask tab, the counts sit behind the chart button (Memory details).

It turns Ready once all four are met: 5 days, 300 frames, 60 distinct things and 3 recurring threads. Readiness follows the weakest of the four, so a lopsided history (many frames but a single day) keeps warming until it fills out.

You can ask questions once the page reads Getting to know you and an AI model is connected. Until then, Ask still shows your saved conversations, and Insights asks you to keep journaling.

In Activity only (Settings: Events only), Everpaper takes no screenshots, so Frames processed stays at 0. The page never reaches Ready, and no Reflection is written. Ask still works once the page reads Getting to know you. To get the Reflection, choose Screen history or Detailed screen history (Settings: Visual history or Adaptive visual history) in Settings → Daily history → History mode.

The Reflection#

With an AI model connected, Everpaper writes a short Reflection: up to three observations about your recent work, each marked with the captured moments it comes from ("From 3 captured moments"). It's written when you open Everpaper and after you end a day. To see it, click the arrow beside the counts at the top of Memory; on the Ask tab, click the chart button (Memory details). A model on this Mac keeps it private: Everpaper Native (macOS 27), oMLX, LM Studio, Ollama or Nativ. See AI on your Mac.

Ask#

Type in Ask a question about your history… and press Return (Shift-Return for a new line). Everpaper answers from what it has kept: your screen history and activity, up to three recorded meetings, and files in folders you added under Settings → AI & summaries → Files you include.

Each answer lists its Sources. Open one to check the original screen, the meeting passage (Transcript · 12:34) or the file. When an answer is a numbered list, each point has Correct (save the right version for future answers) and Forget (stop using that claim).

Conversations are saved: Conversations reopens one, and New conversation starts a fresh one. An empty conversation offers up to three Try asking questions drawn from your own history.

Before an answer is saved, Everpaper checks it against the sources it was given and removes sentences it can't support. The check is strict but not perfect: it can occasionally remove a correct detail or miss an invented name. When an answer matters, open its sources.

Episodes#

Episodes is paused in this beta. It was designed to summarize every ten minutes of activity and group the summaries into six-hour chapters, but this build doesn't write new ones. On a new install the tab stays empty. Your day is still written up in Timeline.

You#

Does this fit? asks you to confirm details Everpaper inferred, such as your role (Yes, that's me / Not me). About you shows what you wrote in Settings → AI & summaries → Your memory & summaries → About you. Who you are is a short portrait your AI model writes from your history; Refresh writes it again. What I'm picking up about you lists traits drawn from your rhythms, the tools you rely on and observations that keep recurring. Confirm one with ✓ or remove it with ✕. How your days have felt charts the mood of recent days.

Insights#

• Worth noticing: patterns rising or cooling across your recent days. Each can become a draft: Draft email… opens a preview you send with Open in Mail, and Draft follow-up event… one you save with Add to Calendar. Nothing is sent or saved until you choose. Drafts don't need AI.
• Worth confronting: up to five frank observations, each shown with its evidence. They appear at the default tone, Balanced, and at Candid. To hide them, choose Supportive in Settings → AI & summaries → Your memory & summaries → Second Brain tone. If your history shows nothing worth confronting, the section stays empty.
• How your picture has changed: how Memory's understanding of you has shifted over time.

Connections#

Your network lists Most connected (the hubs, with how many connections each has), Surprising connections (pairs seen together more often than chance would explain) and Areas (clusters). Show graph draws up to 120 recurring things and their links. When Ask is available, suggested questions under Ask open the Ask tab and ask them.

Click a surprising connection, or a link in the graph, to see why the two go together: how often they were seen together and when last, the captures where both appear (Seen together), and, with an AI model connected, one sentence explaining the link. If the two are really the same thing, choose Merge.

Export#

Export (⌥⌘6) writes files on your Mac; nothing is uploaded. Export Obsidian vault writes one note per day, with its written entry, your notes, a timeline, and the people, sites and files seen that day. Export graph (GraphML + Mermaid) writes your connections as a graph file you can open in yEd or Gephi, plus a Mermaid diagram for Markdown. Reveal in Finder opens the folder. See Export for where the files go and how to use them.

What it learns about you#

As you work, Everpaper reads the captured text and pulls out concrete things: email addresses and phone numbers, files, websites (on sites like GitHub, each repository counts as its own thing) and ticket references like #318. Extraction runs on this Mac on text that has already had secrets removed. Everyday plumbing — sign-in pages, analytics, content networks, and boilerplate files such as a project's readme — is filtered out. Things that keep appearing together become connections.

From what recurs, Everpaper keeps a short list of one-line beliefs: people you deal with, projects, tools and sites, files you return to, rhythms in your work, and open loops (a file or ticket you touched a couple of times and then left for about a week). They're worked out on your Mac without AI, grow stronger when they recur and fade after quiet days. Less certain ones read as (likely) or (tentative). Observations the Reflection keeps making can become traits, which you'll see under What I'm picking up about you.

Memory has no page that lists the beliefs. They shape Ask's answers, Worth confronting and How your picture has changed. To look one up, type in the Index (⌘K): beliefs appear "in the Commonplace Book". You can also list them from the command line. With a model connected, AI refinement of insights (Settings → AI & summaries → Your memory & summaries, on by default) lets the model tidy the list.

How far back it remembers#

On a new install, Everpaper deletes old material after 30 days. In Settings → Privacy → How long it stays, Keep screenshot images for sets the window, and Manage saved text… → Also delete expired text & recall data is on. When a capture passes that window, Everpaper deletes the image, its text, its search entries and the people, files and sites found in it, along with that day's running notes and written journal entry. The notes you typed in the end-of-day review stay. Memory's Frames processed, Topics and Recurring threads, Connections and Ask then reach back only that far. To keep more, choose 90 days or Forever, or turn the switch off: images are still deleted, but the text and what Memory learned from it stay.