Second Brain

The Second Brain is the Commonplace Book in Everpaper: the page that reflects your own work back to you. It reads across every day you've captured, links the things that keep recurring, and forms a small set of plain-language beliefs about what you focus on. You can read those beliefs directly, explore how everything connects, and talk through your history.

Open it from the Index (⌘K) and move between its four sections with the tab bar or ⌘1–⌘4: Chat, Insights, Network, and Export.

Getting to know you

The Second Brain only starts answering once it has seen enough of your work to be useful rather than confidently wrong. The page shows where it stands — Still learning you, Getting to know you, or Ready — with a progress bar and four counts: days, frames, things, and threads.

It turns Ready once all four signals are met:

• 5 days captured
• 300 frames captured
• 60 distinct things seen (people, files, sites, and so on)
• 3 recurring threads — things that show up across more than one day

Readiness is the weakest of the four, so a lopsided history (many frames but a single day) keeps warming until it fills out. Keep journaling and it gets there.

The knowledge graph

As you work, Everpaper reads the captured text on each page and pulls out the concrete, recall-worthy things in it:

• People — email addresses, and recurring phone numbers
• Files — file paths and named files (matched against an extension list, so roadmap.swift is recognized but v1.2 is not)
• Sites — web addresses, kept specific (github.com/example/widgets is its own thing, not collapsed into a bare github.com)
• Projects and tickets — issue references like #318
• Topics — recurring subjects that tie the above together

Each thing is tied back to the exact moment it appeared and timestamped, so it can be ranked and shown without losing where it came from. Extraction runs entirely on your Mac on text that has already been redacted, so secrets such as card numbers, keys, and one-time codes are never turned into entities. Ubiquitous plumbing — sign-in portals, analytics, CDNs, and boilerplate files like README.md or package-lock.json — is filtered out so it never looks like a hub of your work.

When two things keep appearing together, that co-occurrence becomes a link. The result is a cross-day graph of who and what your work threads through.

Beliefs about your focus

From that graph, Everpaper distills a self-model: a set of one-line beliefs about you, each grouped by kind — people you keep dealing with, projects and areas of your work, topics that recur, tools and sites you live in, files you keep returning to, rhythms in how you work, and open loops you touched then dropped.

Beliefs are derived directly from your history, not guessed by an AI, and every one carries an honest confidence marker. A belief with thin evidence reads as (tentative) or (likely); as the same person or file recurs across more days, the belief is reinforced and its confidence grows. The strongest beliefs rise to the top of the Commonplace Book.

A few things keep the picture stable and honest:

• A belief needs to show up across at least two days before it forms.
• Confidence only firms up once enough evidence has accumulated, and a belief that earned trust keeps it.
• A single quiet day can't retire a belief — it fades only after several quiet days in a row.
• A file or ticket you touched a couple of times and then dropped about a week ago is flagged as an open loop, phrased as a possibility rather than a fact.

You can read all of this with no AI provider connected — the beliefs are computed locally and shown as cards.

The Reflection

When you connect an AI provider, the Second Brain gains a deeper layer: the Reflection. After you finish a day — or when you open Everpaper in the morning — it reads the actual moments it captured (not just the counts, but what was on your screen, in what order, at what time) and writes a short Reflection: two or three genuine observations a thoughtful colleague might make.

Not "X is heating up." More like: "You seem to hit a wall around 3pm — switching to email right after compile errors." Every observation points back to the specific moments it noticed it in — if it can't ground an observation in something it actually saw, it doesn't say it.

The Reflection runs on your Mac if you use Apple Foundation Models or Ollama — free, private, no cloud needed. It appears as the lead card in Insights.

How the self-model develops

The Reflection's observations don't just appear and vanish — they become lasting beliefs. When the same pattern recurs across days, the observation is reinforced and its confidence grows. When it doesn't recur, it fades. Patterns that recur enough times promote to stable traits — durable dispositions like "you work best in long morning blocks" or "you avoid documentation until pressure builds."

Those traits then shape what the Second Brain notices next. Its perception is conditioned on what it already understands: if it has learned you're a methodical debugger, it pays attention when you break that pattern — and the breakage is often the most interesting signal. A higher-order reflection pass periodically reads the accumulated beliefs and synthesizes higher-level themes ("your avoidance of docs and tests both point to deferring unglamorous work") and tensions (patterns that pull in different directions).

This is the closed loop: observe → believe → reinforce or revise → perceive differently → observe again. The understanding deepens, not just accumulates. And it's honest at every step: observations are grounded in real captured moments, every chat answer is verified against the evidence, and beliefs that don't hold up decay and retire.

The story of becoming

The "Who you are" card on the You tab synthesizes the self-model into a narrative — not a label, but a story of how the Second Brain's understanding of you is developing. It describes what it's come to understand, how that's changed recently, and where it sees tension:

You're a builder who values deep focus — you work best in long morning blocks, and you've gotten steadier at holding that focus through the afternoon. There's a tension, though: when you hit a compile error, you tend to switch to email — I first noticed this three reflections ago, and it's held.

The narrative is grounded in the accumulated evidence, and it distinguishes what's firm (stable traits) from what's still emerging (new patterns). It updates as the self-model develops — so it tells a story of becoming, not a static portrait.

Insights

The Insights tab surfaces what's worth your attention right now:

• Reflection (when an AI provider is connected) — the lead card, with two or three genuine observations grounded in your recent captured moments.
• Worth noticing — patterns rising or cooling across your recent days, all derived deterministically from your history. Each has a quick action to draft an email or a follow-up calendar event from it.
• Worth confronting — a frank, grounded mirror of things your own history suggests, each shown with its evidence. This stays silent unless your tone dial (in the Colophon) allows it, and it never fabricates a criticism.
• How your picture has changed — the arc of how the Second Brain's understanding of you has shifted over time.

Network

The Network tab is the graph at a glance:

• Most connected — the hubs your work threads through.
• Surprising connections — strong links between things that aren't both central, ranked by how many times more often than chance they appear together.
• Areas — clusters Everpaper detected in the graph.
• Your graph — an interactive node-and-link canvas you can open with Show graph.

The tab also offers a few starter questions; tapping one drops you into Chat with that question already asked.

Chat and recall

Chat is the conversational home of the Second Brain. Ask it about your work — who you keep dealing with, the files and tools you live in, or where your time actually goes — and it answers from what it has captured.

Under the hood, recall grounds every answer in your real history rather than inventing anything:

• It reads your question for what you're after (a phone number, a file, a site) and pulls the matching things from your journal.
• For relational or temporal questions ("how are X and Y connected?", "what have I been doing around Z?"), it assembles real evidence from the graph and answers from that bundle, keeping [#n] citations so you can jump to the exact moment each claim came from.
• Replies are checked against that evidence before they're saved, so a claim the history doesn't support is stripped out. This verification runs on every answer — not just the relational ones — so a fabricated person or project can never slip through.

Conversations are remembered: the Second Brain resumes your thread, keeps a rolling summary so long chats stay coherent, and can open a fresh chat when you want one.

There's also an Explain field for two specific things — enter a person, file, or site on each side and Everpaper shows the shared bridges between them, drawn only from your graph.

Export

Everything stays on this Mac, and Export only materializes what you've already captured:

• Export Obsidian vault — daily notes with backlinked people, sites, and files.
• Export graph — your knowledge graph as GraphML (for yEd, Gephi, Obsidian) and Mermaid (for Markdown).

Both write to your Mac and can be revealed in Finder; nothing leaves your machine.