Read every file

Applies to Canto 1.0 beta 9Updated Oct 27, 2026

On this page 8 sections

Start it
The plan card
On this Mac or a cloud model
While it reads
The answer
What Canto keeps
Settings
Related pages

A chat answer reads a few passages at a time. Ask it about a folder of 387 papers and it can't hold them all: it reads some, and says how many ("Read 13 of 387 files"). Read every file is for those questions. Canto reads each note and file once, keeps what bears on your question with every quote checked word for word against its source, then answers from all of it and says exactly how much it read.

Nothing reads until you press Start reading. Before that, a plan card says what it will read and skip, how long it takes, what it costs and what leaves your Mac.

It's part of Pro. On Free it uses one of the three free tries for deep research on this Mac, spent only when the reading starts.

Start it#

Every way in puts the same plan card in a chat:

From the message field. When what you type asks about all of a set ("go through all of them", "every paper in this folder"), Read every file first? shows above the field (Read every note first? for notes), with how many there are. Read every file… puts the plan card in the chat with your words as its question; Just ask sends your message as it is. It understands English, Korean, French, German, Spanish and Hungarian.
Under an answer that read part of a set: "Read 13 of 387 files for this answer", with Read all 387 and ask again and Research instead.
From a folder. Right-click a folder on this Mac in the sidebar and choose Ask about every file…, or a notes folder and choose Ask about every note…. A #tag page's ⋯ menu has Ask about every note tagged #…, and All notes has Ask about every note…. Each opens a new chat that reads only that place.
From ⌘K: Read every file in a folder… lets you find a folder (or All notes) by name.
The model can suggest it when a chat reads a large set. It shows the same card, and still nothing starts until you press Start reading.

The plan card#

Read every file before answering (or every note, or every note and file) lists:

Row	What it says
Question	What Canto answers from all of them. Edit it before you start.
Reads	How many notes and files, where, how many pages, and about how many tokens. Show the list opens them. Under it, what it skips and why: a scanned PDF while Read scanned PDFs is off (with Open Settings), a locked file, a file in iCloud that isn't downloaded, a file with no text. A file longer than the page limit is read to that page, or in full with Read long files in full.
How	Read in full, or Skim: a paper's abstract, introduction and conclusion, or each section's start for notes. Skim is the default with a model on this Mac.
Read with	The model that reads. It starts as the chat's model; the menu also offers a cheaper one on the same service, with its price, and a model on this Mac.
Answer	Always the chat's own model.

Under them, the estimate: the time, and for a cloud model the cost, both as ranges ("About $18–27 · about 55 min to 1 h 15 min"), with what's left of your monthly limit for cloud models. If the estimate is more than what's left, the card says so, and the reading pauses when the limit is reached. A set of more than 5,000 items is too many: choose a smaller folder or fewer notes.

The last line says what leaves this Mac, and the foot reminds you: Pauses if you quit · Resume any time.

On this Mac or a cloud model#

On this Mac, nothing leaves: the card says "Nothing leaves this Mac. Canto pauses while you chat and when the Mac is on battery." Reading waits while you type or a chat answers, and goes on by itself afterwards. It pauses while the Mac runs on battery or is hot, with Resume anyway.

With a model off this Mac, the text of every note and file it reads goes to that service, a few at a time. Pressing Start reading asks first, every time: Reading these files sends them off this Mac, with what it sends (the text of how many files, and how much), to whom, and what it costs. Send and start reading starts; Read on this Mac instead switches the reader to a model on your Mac.

The answer is a chat turn, so when the chat's model is off your Mac, your question, the chat so far, the claims and quotes Canto kept, and any passage it reads again to check one go to that model. The card says so before you start, even when the reading itself stays on your Mac.

Cloud models are off until you turn them on: see What leaves this Mac.

While it reads#

You can keep chatting. The card shows the progress, the time left, and for a cloud model what it has cost so far, then what's being read now and each finished file's claims and checked quotes ("6 claims · 4 quotes checked"). See all lists every item, with what couldn't be read and why. While it reads, the chat's row in the sidebar shows a ring.

Pause and Stop are on the card. Resume reading carries on where it stopped, in the middle of a long file too.
A busy service is tried again a few times, then the reading pauses. Reaching your monthly limit pauses it at once, with Raise the limit….
Quitting pauses it. The next time Canto opens, the card says "Paused when Canto quit", and nothing resumes until you press Resume.
It stays inside what the chat reads. Narrow the chat's sources, or turn folders on this Mac off, and the files that left are taken out of the reading and its answer.
A paused or stopped card can also Answer from what's read, Read the rest on this Mac or Choose another model.

One chat reads one set at a time, and a reading waits for the chat's follow-up queue to finish (and the queue for the reading).

The answer#

When the reading finishes, the chat's model writes the answer as a normal chat answer, under a line that says what was read:

"Read 78 of 78 notes and 380 of 387 PDFs · 7 couldn't be read · 1 read to page 300", with What wasn't read,
then who read them, the tokens, the cost and the time.

Each file is cited with the page number printed on it, and a citation opens the PDF at that page. How I got this says how many were read one by one and with which model, how many bore on the question, how many quotes were checked word for word (and any that couldn't be found, which weren't used), and what wasn't read, by reason.

What Canto keeps#

What Canto read stays on this Mac. It's never synced, backed up or exported, so you can ask more about the same set without reading it again.

Beside the chat, the inspector's Context has Read for this chat · kept on this Mac: the counts, a search field (Search what Canto read) and each file's claims and quotes. Forget what Canto read… removes it; the answer stays, and asking about those files again reads them again.

On another Mac, the chat shows the answer and which Mac read the files: "Read on MacBook Pro · 380 of 387 · the details stay on that Mac".

Settings#

Settings › Assistant › Reading every file:

Tell me when reading finishes (on): a notification when Canto has read every file and no Canto window is in front.
Long files (Read to page 300): a longer file is read to this page unless you choose to read it all. Also 100, 600 or 1,000.
Pause on battery (on): reading on this Mac's models waits while the Mac runs on battery.

Ask about your notes: choosing what a chat reads.
Research and research notes: a planned, cited report from your notes and the web.
Servers and cloud models: the models that can read off your Mac, and the monthly limit.
Free and Pro: free tries and what Pro includes.