In this catalogue
Six things
the chatbots
can't do.
Productivity professionals don't need another chat window. They need a system that captures, structures, and surfaces — so the assistant arrives at every conversation having done the reading. Moonjar is that system.
01 — Capture
Save anything,
from anywhere.
The hardest part of building a knowledge base isn't AI. It's the moment between I should save this and it's saved. Moonjar collapses that moment into one tap, one paste, or one forwarded email.
Type a thought. Paste a URL — the article is fetched, parsed, and stored, not just bookmarked. Snap a photo of a whiteboard, a receipt, a page from a book. Drop in a PDF. Forward any email to your private capture address and Moonjar treats the body and any attachments as a single document, sender verified.
From the share sheet, every iOS app becomes a save button — Safari, Mail, Photos, anything you can share goes straight into the library.
- TEXT"Take a closer look at the Hass-Hadar paper."09:12
- URL stratechery.com/2026/agents-and-the-…09:18
- PHOTOWhiteboard at standup · 1.4 MB09:46
- PDF Q4-board-deck.pdf · 3.1 MB10:03
- EMAILFrom legal@…/ Lease v3 · 1 attachment10:21
- VOICE"Ask Sarah about the lease before Friday…"11:04
ARTICLE · auto-classified
Building beyond the chatbot
- Author
- Ben Thompson
- Source
- Stratechery
- Saved
- Today, 09:18
- Reading time
- 14 min
- Topics
- AI strategy · agents · retrieval
- Summary
- Why the next wave of AI products won't look like chat — and what that means for incumbents.
RECEIPT · auto-classified
Bunnings — 14 Apr 2026
- Total
- $184.20
- Merchant
- Bunnings Warehouse
- Category
- Home · garden
- Tax
- $16.74 GST
02 — Structure
Knowledge that knows
what it is.
A note app gives you a stream of text. A folder app gives you a stream of files. Moonjar gives you typed, structured records.
Every captured thing is read on arrival, classified into one of eleven types — article, receipt, contact, recipe, contract, photo, transcript, and so on — and the right fields extracted. Articles get authors and summaries. Receipts get amounts, merchants, tax. Contacts get emails and roles.
The difference matters when you go looking. You can search by what something says. You can also filter by what it is: receipts over $100 from last quarter; articles by a specific author; contacts at a specific company.
03 — Search
Hybrid recall
across everything.
"Search my files" is a checkbox feature in every assistant. It is also, in practice, useless past a few hundred documents. Embeddings drift; keywords miss; long context costs a fortune.
Moonjar runs three searches at once: vector similarity over semantic embeddings of every document, full-text search across the words you actually wrote, and structured field queries against the typed metadata. The results merge intelligently. Recall is high. Latency stays in the tens of milliseconds.
The result is a system that finds the receipt from a contractor you haven't named in two years, the article you remember was about something to do with attention heads, and the meeting note where someone mentioned the Q3 number — all from the same query.
what was that piece on attention sinks
- ARTICLE StreamingLLM: attention sinks & cache arxiv.org · saved 8 Mar semantic · 0.91
- NOTE Meeting w/ Sarah — KV cache notes 7 Mar · "…attention sink trick…" full-text
- PDF Inference scaling — internal memo 22 Feb · 14 pages semantic · 0.78
What did Sarah say about the lease last week?
From your meeting note on 14 Apr: she's pushing for a 7-year term with a CPI cap, but is open to a 5+5 break clause if we agree on a fitout contribution. Her email the next morning attached Lease v3.pdf — clause 12.3 reflects the CPI cap.
3 sources · open
Remind me to send our redline by Thursday 5pm.
04 — Chat
An assistant that's
done the reading.
Before you've finished typing, the chat has already searched your library, opened the relevant documents, and consulted its memory of past conversations. By the time it answers, it's not improvising. It's reporting.
It can also act. Capture a new note. Add a row to a collection. Set a reminder. Generate a one-page brief. Share an artifact as a link. The same assistant that knows your stuff can do something with it.
And memory persists. Mention a colleague's name once; it sticks. Tell it your son's birthday; it sticks. The next conversation starts from a base of context, not a blank page.
05 — Collections
Build your own
database, by talking.
Some things you save aren't documents — they're rows. Books you've read. Wines you liked. Companies you're tracking. People you owe a reply. A spreadsheet would do, except spreadsheets need maintaining.
In Moonjar, you describe what you want to track in plain language: "track every book I've finished, with author, rating, theme, and the one line I want to remember." The assistant builds the schema, populates it from anything relevant you've already captured, and adds a button to add more.
Collections cross-reference each other. The book you're reading can link to the author. The author can link to other books. None of this requires a schema editor — only a sentence.
| Title | Author | ★ | Theme |
|---|---|---|---|
| The Beginning of Infinity | D. Deutsch | 5 | Epistemology |
| Working in Public | N. Eghbal | 4 | Open source |
| The Years | A. Ernaux | 5 | Memoir |
| Algorithms to Live By | B. Christian | 4 | Decisions |
| + Add row · or just say "I finished <book>" | |||
Forward anything. Verified senders only.
06 — Everywhere
Where your work
actually happens.
The assistant lives in five places, because thinking does. The phone catches the idea on the train and the voice note on the way home. The iPad is where you read long things and ask the assistant about them. Email is where everyone else sends you stuff. The share sheet is the universal "save this for me" button. The watch is where you check today's pile and the alerts that need you.
What you capture on iPhone, iPad, email, or the share sheet is on every other surface in seconds. Push notifications nudge you when a saved document expires. The whole thing syncs without you noticing.
And quietly, in the background
A handful of things that aren't headline features but you'll notice on day three.
- · Voice transcription and read-aloud playback
- · Daily notes — append-only journal that the assistant can read
- · Reminders with date/time scheduling and push delivery
- · Automations: scheduled scanners that surface stale documents and forgotten threads
- · Public artifact links — share a generated brief without a sign-up wall
- · Built-in tools: weather, web search, places, routes, news, flights
- · Push for document expiry and reminder warnings
- · Per-document share links with rich previews
Colophon
If your day involves more reading and remembering
than the chat window can hold —
Moonjar is the surrounding system. Capture is fast. Search is comprehensive. The assistant is genuinely informed.