The problem with a workspace is rarely storage. It is that six months later you can describe the thing you need and cannot name the file it is in. Most tools answer half of that: a notes app searches your notes, a drive searches your files, and the answer is in whichever half you did not search.
- Board memo — Q3 · Notes · edited 2 days ago
- Board-memo.lumendoc · Library / Finance · 1 match on p.3 — “…the board memo should carry the variance, not the raw total…”
- Q3-model.lumensheet · Library / Finance · linked from the memo
- Q3-board-pack.pdf · Library / Board · 1 match on p.12
Where the line is
“Local-first” is a claim worth being precise about, because most products that make it mean something weaker. Here is every kind of finding you do in LumenQube, and whether it crosses the line between this machine and the network.
| What you are doing | Where it happens |
|---|---|
| Full-text search over notes and documents | This machine — works with the network off |
| Wiki-links and backlinks | This machine — works with the network off |
| Opening, editing and exporting | This machine — works with the network off |
| Notes at rest | Encrypted on this machine, with your OS keychain |
| Semantic search — asking in your own words | Cloud. Needs an account and a connection |
| Asking the agent a question about a document | Cloud model, on the passages it needs |
| Transcribing a saved recording | Managed service — the audio is uploaded |
Three different ways of finding
You remember a word that was in it
Full-text search. Instant, on your machine, offline, and it reads the inside of a PDF rather than only its filename — a match on page 12 tells you it is on page 12.
You remember what it was connected to
Write
[[like this]]to link one page to another, and the page you linked to lists what points at it — carrying the sentence the link appeared in. A connection you made by hand survives being forgotten, and you can walk it from either end.You can only describe it
Semantic search finds the thing you can describe but cannot name — “the thing we decided about pricing in the spring”. This is the one search that leaves your machine: the meaning-matching runs in the cloud today, and it needs an account.
A to-do list that agrees with your notes
Tick a box in a page and it is the same commitment everywhere it appears. Tasks are tracked by identity, not by their position in a document, so reordering a page does not scramble what is done — and the same task shown in two places is one task, not two rows that happen to match.
What the agent remembers, and what it does not
LumenAgent keeps durable memory across sessions — preferences, context about your work, things you have told it once and should not have to repeat. Credentials are screened out before anything is stored. But that memory has no source links, so it remembers what you told it rather than the decision you made in a document three weeks ago, with a citation. Search is the half that has the citation.
Questions about search
Does search work with the network off?
Full-text search does, over both your notes and your document library, and so do wiki-links and backlinks. Semantic search does not — it is the one search that needs an account and a connection.
Does it search inside PDFs and spreadsheets, or just filenames?
Inside them. A result names the page or the location of the match, so a hit on page 12 of a board pack tells you it is on page 12.
Are my notes and my documents two separate searches?
No — one box, one result set, with counts you can filter by. That is the point of the feature: the answer is usually in whichever half you would not have searched.
Is anything indexed in the cloud?
The full-text index is built and queried on your machine. Semantic search sends the query, and the passages it needs to compare, to a cloud model — which is why it is a separate, account-gated feature rather than the default.
What can I read next?
The Search page carries the same table as a live demonstration, and what actually leaves your machine traces two real requests end to end.
