VaultBook AI
Your notes remember
Your notes remember
so you don't have to
VaultBook AI doesn't live on a server. It lives inside your library — learning your rhythms, surfacing what matters, and connecting ideas you forgot you had. No cloud, no prompts, no waiting. Just your second brain, always on.
🧠
Thinks like you do
Learns your reading patterns, your weekday habits, your preferred tools — and quietly adapts.
🔎
Finds what you mean
Ask in plain language. The AI searches titles, text, OCR, attachments — and ranks by what matters most to you.
🔒
Never leaves your device
Every AI feature runs locally in your browser. No API calls, no data sent anywhere. Your intelligence, your machine.
AI that doesn't interrupt — it anticipates
Most AI tools demand your attention. They want you to type a prompt, wait for a response, and figure out if the answer is useful. VaultBook does the opposite. Its intelligence is ambient — woven into the fabric of the workspace so deeply that you stop noticing it's there. You just find the right note faster, see a connection you'd missed, or discover that the entry you need is already waiting at the top of the page.
That's the design principle: AI should match the speed of thought. Not slower. Not louder. Not somewhere else. Right here, in the flow of your work, as fast as your fingers can move.
Two engines. One workspace.
The Suggestions pager anticipates what you need next. Smart Search finds everything else.
✨ Suggestions Pager
It knows your Monday
is different from your Thursday
is different from your Thursday
Open VaultBook and the Sparkle pager already knows what to show you.
It analyzes the last 4 weeks of your reading patterns for the current day of the week,
surfaces upcoming scheduled entries, and ranks your recently read notes, files, and tools
by personalized relevance — not recency, not alphabetical order, but how
you actually use your library.
4-page carousel: Suggestions, Recently Read, Recent Files, Recent Tools
Top 3 entries for today's day-of-week over the last 4 weeks
Upcoming scheduled entries surfaced within a 48-hour window
Personalized relevance distribution learned from your usage
🔎 Smart Search & QA
Ask a question.
Get the answer, not a list.
Get the answer, not a list.
Type a question in natural language and VaultBook searches across everything
you've ever written — titles, body text, labels, sections, inline OCR, attachment
contents — with weighted scoring that puts the most relevant answer at the top.
Not ten blue links. The answer.
Weighted scoring: titles (8×), labels (6×), OCR (5×), body (4×), sections (3×)
Paginated results (6 per page) with prev/next navigation
Typeahead dropdown with real-time suggestions as you type
Past query suggestions for instant recall of previous searches
💡 Related Entries
The connection you didn't know existed
When you open a note, VaultBook quietly scans your library for contextually similar entries
and surfaces them in a fade-in panel. It's the moment where a project brief links to a
research note from three months ago, or a meeting summary connects to a draft you never finished.
Upvote the good matches, downvote the noise — your votes persist across sessions and the
relevance model reshapes itself around your judgment.
Contextual similarity computed across your full library
Reddit-style upvote/downvote per pair
Votes persist in repository and retrain relevance
Paginated with smooth fade-in/out animation
It reads what you can't type
Photos, scans, PDFs, spreadsheets, presentations, emails — VaultBook extracts text from all of them and makes it searchable.
Inline OCR
Every image becomes searchable text
Paste a screenshot into a note. Attach a photo of a whiteboard. Drop in a scan of a receipt.
VaultBook automatically runs OCR on every inline image, caches the extracted text per entry,
and indexes it for search. The next time you search for a word that only exists inside a photo,
VaultBook finds it — instantly, without you ever having to transcribe anything.
The top 12 QA results also trigger background OCR warm-up if it hasn't been done yet,
so search results get smarter the more you use them.
Deep PDF indexing
Text layers are extracted from attached PDFs. Scanned pages are OCR'd — so even a photographed contract is fully searchable.
Office document parsing
XLSX, XLSM, PPTX, and DOCX files are cracked open and their text content indexed. Embedded images inside these files are OCR'd too.
ZIP & MSG archives
Attach a zip file or an Outlook .msg email and VaultBook indexes every text-like file inside — plus deep-indexes attachments within those emails.
It learns by watching you work
No training step. No configuration. VaultBook's AI gets sharper every time you search, read, or vote.
Vote-Based Learning
Your judgment reshapes the results
Every time you upvote a search result or a related entry, VaultBook boosts it in future rankings.
Every downvote pushes noise to the bottom. These votes persist across sessions in your repository,
so the system compounds your preferences over weeks and months. The confirmation dialog teaches you how
voting works the first time, then gets out of your way. An undo toast means nothing is permanent.
Over time, the workspace stops showing you what's technically relevant and starts showing you
what's actually useful — a distinction no static algorithm can make.
Smart label suggestions
When you edit a note, VaultBook analyzes its content and suggests labels as pastel-colored chips — so tagging is a single click, not a memory exercise.
Query memory
Past searches are remembered and suggested the next time you start typing, so recurring questions are answered in two keystrokes.
Attachment warm-up
When you search, the top 12 results trigger background text extraction on their attachments — so the next search across those files is faster and deeper.
What changes when AI lives in your workspace
Not a chatbot you visit. A layer of intelligence across everything you already do.
Without VaultBook AI
Scroll through dozens of notes to find the right one
Manually tag every entry and hope you're consistent
Screenshots and scans are unsearchable dead weight
Forget what you worked on last Tuesday
PDF and spreadsheet content locked inside files
Search returns everything or nothing
With VaultBook AI
The right note surfaces before you search for it
Labels suggested automatically from your content
Every image OCR'd and indexed — searchable instantly
Weekday patterns show what you were focused on
Text extracted from PDFs, XLSX, PPTX, ZIP, and emails
Weighted results that improve with every vote
AI that respects the boundary
Every computation happens in your browser tab. Nothing is sent, stored, or processed anywhere else.
Privacy Architecture
Zero network. Zero trust required.
VaultBook AI is not a wrapper around a cloud API. There is no server. There is no model endpoint.
Every feature — OCR, text extraction, relevance scoring, vote-based learning, pattern
analysis, label suggestions — runs entirely inside your browser using local computation.
Your notes, your votes, your reading patterns stay in a JSON file on your disk.
If you disconnect your Wi-Fi and keep working, nothing changes. That's the test.
No API keys
There's no OpenAI key to configure, no usage limits to monitor, no API bill at the end of the month. VaultBook's intelligence is self-contained.
Your data format, your data
Notes are stored as Markdown sidecar files. Votes and metadata live in a single repository.json. Everything is portable, human-readable, and yours to move.
Works offline forever
VaultBook doesn't phone home for updates, license checks, or feature flags. Once downloaded, it runs without a network connection — indefinitely.
Ready to think faster?
Download VaultBook and let your workspace start learning how you work.