Guide
Local-first software stores your data on your own device as the primary copy, keeps working offline, and treats any cloud as optional. It matters most for sensitive data like screen history. Dayflow is one example: recordings and the timeline stay on your Mac, and its Local AI mode keeps screenshot analysis on-device through Ollama or LM Studio. A cloud provider or Pro sends selected screenshots off-device.
You type something private into an app. A journal entry, a note about a doctor’s appointment, a list of what you actually did on a Tuesday when you were supposed to be writing a report. Later, a small question shows up: where did that sentence go? Is it on your laptop, on a server in another country, or both? If the company raises prices, gets bought, or closes, do you still have it?
Most people never find out. The app opens fast, sync works, and the question only comes back when an email arrives about updated terms, or a login fails and years of notes are on the other side of it.
This guide explains what “local-first” means in plain terms, why the location of personal data matters, and seven checks you can run on the apps you already use. Most of them need nothing more than a sheet of paper and a few quiet minutes. Disclosure: Dayflow publishes this article, and Dayflow is our product. It comes up once, in its own section, along with the cases where it is the wrong choice. Everything said about it comes from its own documentation and published policy as reviewed on 2026-09-06, for version 2.2.0.
The phrase gets used loosely, so here is the plain version. Software is local-first when:
Two nearby terms are worth separating. Cloud-first software keeps the real copy on the vendor’s servers and shows you a window onto it. Offline-only software never connects to anything. Local-first sits between them: connected when useful, but never dependent on the connection.
It is also worth knowing that “local” is now a marketing word. An app can store files locally and still send their contents out for a feature you clicked once and forgot about. The label is a starting point for questions, not an answer.
A few plain reasons, all familiar from everyday experience:
Every copy is a place it can be read. Data on a server is also in that server’s backups, its logs, and possibly its support tools. Each copy has its own access rules. Data that never leaves your laptop has one copy to worry about.
Your access depends on someone else’s decisions. With cloud-first software, your notes are reachable only while your account is active, the service is running, and the price is one you are willing to pay. Many people have had the experience of a favorite app shutting down with a few weeks’ notice.
Terms change after you upload. Data you sent under one privacy policy lives under the next one too.
Storing and processing are different. This is the one that catches people. An app can keep your files on your computer and still send their contents to an outside AI model every time you ask for a summary. Where data sleeps and where it gets read are separate questions.
Personal data adds up. One note is minor. Years of notes, or months of screen history, describe a life in detail. The more of it there is in one place, the more the location matters.
Local is not automatically safe. Keeping data on your own machine moves responsibility to you. A lost laptop without disk encryption, or a dead drive with no backup, is its own kind of data loss. Local-first is a trade, not a free lunch.
What to do: Take a sheet of paper and draw three columns: the app, what personal data is in it, and where the main copy lives. For that third column, write “my computer,” “company server,” “both,” or “don’t know.” Fill it in from memory first. Then, for every “don’t know,” spend two minutes in that app’s settings or help pages looking for the answer.
Illustration: Someone lists their notes app, their journal, their photo library, a running work log, and a document with account recovery codes. From memory, they mark three as “don’t know.” After checking, they find the journal entries exist only on the company’s server, and the recovery codes are in a document that syncs to an account they rarely use.
After a week: The “don’t know” column should be short or empty. A good test: for each sensitive item on the list, you can say in one sentence where it is stored. If you can, the map has done its job, and you know where to focus the rest of these checks.
What to do: Turn off Wi-Fi for 15 minutes. Open each app from your map. Try three things in each: read something old, edit something, and search. Note which apps behave normally, which show a spinner or a login screen, and which show partial data.
Illustration: The notes app opens and searches fine. The journal shows a blank page with a “reconnecting” message. The calendar shows this week but nothing from last year. That tells the person the notes are truly local, the journal is not, and the calendar holds only a short cache.
After a week: You know which apps hold a real local copy versus a temporary one. If the most important item on your map failed the test, you have a clear decision to make: accept the dependency knowingly, or move that data somewhere that passes.
What to do: For each important app, find the export option and export one month of data. Then open the exported file with the most basic program you have, such as a plain text editor. Is it readable? Plain text, Markdown and CSV files can be opened by almost anything for a long time. A single file only the original app understands is a weaker guarantee.
Illustration: One app exports a tidy folder of Markdown files, one per entry. Another produces a single archive that will not open without the app. A third has no export at all, only a “contact support” link. The person now knows which data they truly control.
After a week: You have at least one readable export of each important thing, stored on a drive you own. Put a monthly reminder on your calendar to repeat it for anything that cannot sync to a local copy on its own.
What to do: You do not need to read a whole privacy policy. Open it and search for four words: “retain,” “delete,” “train,” and “third.” Read the sentences around each one. Then write one line per app on the back of your map, such as “deleted after processing” or “kept while the account is active.”
Illustration: One policy says uploaded files are deleted after processing, but usage records are kept for sync. Another says content may be used to improve the service. A third says nothing clear at all. Each of those is a different promise, and now they are written down side by side.
After a week: You can state the retention rule for each app in one sentence. Anything that says your content may be used to train models, or that is vague about deletion, is flagged so you can decide what you are comfortable putting in it.
What to do: For every AI feature you use, ask one question: when I press this button, where does my content go? Look in settings for wording like “on-device,” “local model,” or the name of a cloud provider. Treat chat assistants the same way. Anything you paste into one, or allow it to retrieve, goes to whoever runs it, no matter where the original file lives.
Illustration: A person’s notes are stored locally, but the “summarize” button sends the note text to an outside model. Their writing app has a local option that is turned off by default. They switch it on for the private notebook and leave the cloud option for work that is already public.
After a week: Every AI feature you use is marked on your paper as either “on-device” or “sent to a provider.” The features you are not comfortable with are off, and the ones you kept, you kept on purpose.
What to do: Sort your personal data into three piles on paper. “Fine in the cloud” might be shopping lists and shared calendars. “Cloud with care” might be work documents. “Never” might be health notes, financial details, private journal entries, or a record of everything on your screen. Then compare the “never” pile with your map from the first step.
Illustration: A person puts their private journal in “never” and notices it is the same journal that failed the Wi-Fi test. That match, not a vague feeling about privacy, is what prompts them to change it. For their work record, they decide paper is enough, and print a daily work log template that never goes near a network.
After a week: Nothing in your “never” pile sits in an app marked “company server.” Where something still does, you have chosen that tradeoff with eyes open.
What to do: Local-first means you are the backup plan. Pick one external drive you own and copy your important local data to it. Make sure full-disk encryption is turned on in your Mac’s settings, so a lost laptop is not also a privacy breach. Write the backup date on a sticky note on the drive.
Illustration: Someone moves their journal to a local app, feels good about it, and then realizes it now exists in exactly one place. A ten-minute copy to a drive in a desk drawer fixes that.
After a week: You can name where the second copy of your most important data is, and when it was last updated. If the answer is “nowhere,” the move to local data is only half done.
Screen history is about as personal as data gets. It can show messages, documents, bank tabs and half-written emails. That makes it a good test case for every check above, and it is the kind of record Dayflow keeps.
Dayflow is a free, MIT-licensed Mac app for macOS 14 or later. It captures your screen at one frame every 10 seconds and has AI write the day into a readable timeline, so you do not have to start a timer or reconstruct where your day went from memory.
Run it through this guide’s questions:
More detail is in local and cloud processing and is Dayflow safe?
Dayflow publishes this article, and Dayflow is our product.
If the worry behind your search is being watched rather than where files live, time tracking without surveillance is a closer match.
Local-first is a simple idea: the real copy of your data lives with you, and everything else is optional. It matters most for the data that would hurt to lose or to have read, and it shifts some responsibility, like backups and disk encryption, back onto you.
You do not need new software to start. A sheet of paper, 15 minutes with the Wi-Fi off, one export, and four words searched in a privacy policy will tell you more about your apps than any label on their websites. Then decide what belongs in your “never” pile, and make sure your tools agree with you.
Local-first software keeps the main copy of your data on your own device. The app opens, edits and searches that data without a network connection, and any cloud sync or backup is a helper rather than the place the data really lives. You can usually export your data in a readable format, and you keep access to it even if the company behind the app changes course.
No. Offline-only software never connects to anything. Local-first software can still sync between your devices, share with other people, or use cloud features, but your device holds the primary copy and the app keeps working when the connection drops. The difference is which copy is the real one. In local-first apps, it is the one on your computer, not the one on a server.
Often, but not automatically. Keeping data on your device means fewer copies on other people's computers, and no company can read what it never receives. But a local app can still send data out for AI features, analytics or sync, and a stolen laptop without disk encryption is a privacy problem too. Check where data is stored, where it is processed, and how your device is protected.
Not necessarily. Many apps store data locally but send some of it elsewhere for specific features, such as AI summaries, sync, or crash reports. Read the settings for each feature, not just the storage claim. Look for wording like on-device, local model, or the name of a cloud provider. If a feature sends content out, the provider's own terms apply to what it receives.
Yes, if the model runs on your own computer. Tools such as Ollama or LM Studio let a supported model run locally, so the content being analyzed never goes to an outside server. The tradeoff is hardware: local models use your own processor, memory and storage, and setup takes more effort than a cloud option. Cloud AI features send the content to whoever runs the model.
Turn off Wi-Fi and open the app. If you can still read, search and edit your data, there is a real local copy. If it shows a spinner, a login screen or a blank page, the main copy is on a server. Then look for an export option and try it once. A readable export file is a good sign you control the data.
Yes. Dayflow stores full screen recordings and its local timeline on your Mac. What happens during analysis depends on the mode you pick. Local AI keeps screenshot analysis on-device. Your own cloud provider or Dayflow Pro processes selected screenshots off-device. Under the privacy policy, Pro deletes screenshots after processing and may retain Activity Data for sync, while your own provider has its own terms.
No. A paper notebook beats it on privacy. A handwritten log contains only what you choose to write down, never touches a network, and cannot record a bank tab or private message by accident. Dayflow records your screen, which captures far more than you would ever write. Its advantage is that you do not have to remember to log anything, but if privacy is the top priority, paper wins.
The files on your computer stay there, which is a large part of the appeal. Whether they stay useful depends on the format. Plain text, Markdown and CSV files open in almost anything for years. A proprietary database may need the original app to read. Export your important data in a readable format now, while the app still works, and store it somewhere you own.
Under Dayflow's privacy policy, Dayflow Pro deletes screenshots after processing and may retain Activity Data for sync. Pro costs $20 a month and handles cloud analysis for you. If you do not want any screenshots processed off your Mac, use the Local AI mode instead, which runs a supported model on your Mac through Ollama or LM Studio and keeps screenshot analysis on-device.
Dayflow is $0 and open source. Full recordings stay on your Mac, with local AI available for on-device analysis.
Download for MacOS macOS 14+ · MIT licensed · no account required