Guide
Ask six things: where frames are stored, where they are analyzed, what leaves the device when you chat with them, who can open them, how long they are kept, and whether you can read the code. Windows Recall wins on Copilot+ PCs, screenpipe wins if you need audio or Linux, and Dayflow fits Mac users who want an MIT-licensed record with local AI.
You find an app that promises to remember everything you did today. You would never again have to dig through browser history for that one article, or rebuild your afternoon from memory. Then you picture the afternoon itself: the bank tab, the message from your sister, the medical portal, the coworker’s private DM, the password you briefly pasted into a note.
That uneasy pause is reasonable. A tool that sees your screen sees more about you than almost anything else on your computer. The question is not whether to feel cautious. It is what to check so the caution turns into a decision.
This guide covers why screen-reading AI feels different from other software, six questions to work through (most of them with paper and a pen), and how three real tools answer them. Dayflow is our product, so weigh that. The comparison is based only on each vendor’s own published documentation, checked in early October 2026.
Most apps see a slice of your life. A calendar knows your meetings. A browser knows your tabs. The screen is where all of those slices meet, along with things other people sent you in confidence. A record of the screen is closer to a diary written by someone looking over your shoulder.
It also tends to capture things you never chose to save. You might keep a document for years on purpose, but a two-second glance at a bank balance was never meant to be stored anywhere. Screen memory turns passing moments into data.
The risk does not come from one place. Many people picture a single bad outcome, like “the company sees my stuff.” In practice there are several separate points where data can go somewhere you did not expect:
A tool can be careful at one point and loose at another. That is why “is it private?” rarely has a yes or no answer, and why the questions below take each point in turn.
What to do. Before reading a single privacy policy, find out what you would be protecting. Keep a sheet of paper next to your keyboard for one normal workday. Every time something sensitive appears, write one line: the time, the app, and what it was. Do not judge it, just log it. If you like structure, the daily work log template works fine printed out.
Illustration. A freelance designer’s list might read: 9:40 client contract with rates, 11:15 personal banking, 1:30 a friend’s message about a health scare, 3:00 password manager unlocked, 4:20 a client’s unreleased product mockups. Three of those five belong to other people.
After a week. You should have a short list of apps and sites you would never want recorded, and a sense of how often they appear. If the list is long and scattered across the day, you need a tool with strong pause or filter controls, or you need to keep sensitive work on a different device. If it is short, a simple “pause before banking” habit may be enough.
What to do. Draw a small map on paper with three boxes: your computer, the vendor, and any other company. Then read the tool’s privacy page and draw an arrow every time data moves between boxes. Do this separately for storage (where the recordings live) and analysis (where the AI looks at them). If you cannot draw the arrows from the documentation, that tells you something too.
Illustration. One tool might keep recordings on your disk, but its default AI mode sends selected screenshots to a cloud model. Your map has an arrow from your computer to any other company even though the recordings themselves never leave. Another tool might do both storage and analysis on the device, with only small exceptions, such as occasionally fetching a website’s icon from the internet. Both could reasonably call themselves “local.” Your map shows the difference.
After a week. You should be able to explain to a friend, in two sentences, where your screen data goes. If you still cannot, you have not found the answer yet, so do not turn recording on.
What to do. This is the step people skip. Even when recordings and analysis stay on your machine, chatting with a cloud assistant about your history can send that history to the assistant’s provider. Write down which assistant you would use and where it runs. Then read whether the tool says what an assistant receives when it answers you.
Illustration. You ask, “What did I work on with the Henderson account last Tuesday?” To answer, the tool retrieves Tuesday’s records and hands them to the assistant. If that assistant runs in the cloud, Tuesday’s records just left your Mac, even though they were stored locally a second ago. Local storage does not make a cloud conversation local.
After a week. Look back at the questions you asked. If any of them pulled in material from your step 1 list, decide whether you are comfortable with that or whether you should switch to a model that runs on your own hardware.
What to do. Walk around your actual situation, no software needed. Does anyone else ever use your computer? Does it lock quickly when you step away? Is it a work laptop your employer owns and manages? Write down every person who could plausibly sit at the keyboard while you are logged in. Then shorten your screen lock delay and, if the Mac is shared, give each person their own user account.
Illustration. A parent working from home leaves the laptop open on the kitchen table. A teenager borrows it to check something. Without a separate account or a quick lock, the teenager has the same access to a screen history that the parent does. Some tools add their own gate. Others rely entirely on your computer’s login.
After a week. Count how many times you walked away without locking. If the answer is more than zero, a screen history raises the stakes of that habit. For work devices, the honest rule is simple: the employer owns the hardware, so keep personal life off it. If this worries you more broadly, the piece on time tracking without surveillance goes deeper.
What to do. Decide on a retention window before you start, based on what you want the record for. Writing a weekly status update needs about a week. Filling in a timesheet at month’s end needs about a month. Write your number down, then add a recurring calendar reminder to review and clear anything older.
Illustration. A developer wants screen memory mainly to answer “what was I doing before lunch?” and to write Friday updates. Two weeks covers that with room to spare. Keeping six months of frames adds risk without adding much use. Some tools set storage limits and delete the oldest snapshots first when the limit is reached. That is a disk safeguard, not a privacy choice you made.
After a week. When the reminder fires, notice whether you ever looked at anything older than your window. If you did not, your number is right or could be shorter.
What to do. Find two documents: the privacy policy and the software license. For the license, look for a file named LICENSE in the tool’s code repository, if it has one. Ask three things. Can anyone read the source code? Can I use this for my job under its license? What does the company sell, and does that give it a reason to want my data?
Illustration. One tool publishes its code under the MIT license, which lets anyone read, audit and reuse it. Another publishes its code but under a commercial license that allows personal use and requires payment for business use. That is still readable code, but “source-available” and “open source” are not the same promise. A third publishes no code at all and asks you to trust its documentation.
After a week. You should know whether you are allowed to use the tool for paid work, and whether anyone outside the company could check its privacy claims. If either answer surprises you, that is worth knowing before your history grows.
All of this caution is worth it only if the record gives you something back. For many people, that something is not having to remember. You do not have to start a timer, and you do not have to rebuild a lost afternoon from scraps of browser history, a problem covered in where did my day go.
Dayflow records the screen at one frame every 10 seconds and has AI write the day into a readable timeline. Full screen recordings and that timeline are stored on your Mac. You choose where analysis happens: Local AI runs a supported model on your Mac through Ollama or LM Studio, so screenshot analysis stays on-device. Your own cloud provider receives selected screenshots under its own terms. Dayflow Pro manages cloud analysis and, under the privacy policy, deletes screenshots after processing and may retain Activity Data for sync. The local and cloud processing page lays out each path.
If you ask an assistant about your day, it can receive the Dayflow data it retrieves in that conversation. You can export a date range as Markdown from Settings, then Export. Dayflow is $0 and MIT licensed, with an optional Pro plan at $20 a month, for macOS 14 or later. AI summaries can miss or misinterpret work, so check them before sharing. More detail is in is Dayflow safe.
Dayflow publishes this article.
Be direct with yourself about fit. Dayflow is the wrong choice in several cases.
You work on Windows. Dayflow runs only on macOS 14 or later. On a Copilot+ PC, Windows Recall is the tool that applies. Microsoft states it is opt in, stores snapshots locally and always encrypted, analyzes them on the PC’s neural processing unit, requires Windows Hello to open, and filters passwords, national ID numbers and credit card numbers by default. Microsoft also notes that some diagnostic data may be sent depending on privacy settings.
You want audio, meetings or Linux. Dayflow sees only the screen. screenpipe captures screen and audio, transcribes system and microphone audio with speaker identification, stores data on your device by default, and runs on Mac, Windows and Linux. Its cloud options can send data off-device, and commercial use requires a paid license.
You want something to stop you, not show you. Dayflow is a record. It is not a coach, it does not block websites, and it does not produce invoices, payroll approvals or team reports.
| Question | Dayflow | Windows Recall | screenpipe |
|---|---|---|---|
| Platforms | macOS 14 or later | Windows 11, Copilot+ PCs only | Mac, Windows 10/11, Linux (x86_64) |
| What it captures | Screen | Snapshots of the active screen; Microsoft states no audio or continuous video | Screen and audio |
| Where it is stored | On your Mac | Locally, always encrypted | On your device by default |
| Where it is analyzed | On-device with Local AI; off-device with your own provider or Pro | On the PC’s neural processing unit | Local models, or cloud options that send data off-device |
| Code and license | MIT | Not open source | Source-available, Screenpipe Commercial License; older releases MIT |
| Price | $0; optional Pro $20 a month | No published price; see the vendor’s page | Free $0; Basic $21/month or $250/year billed annually; Business $42/seat/month or $500/seat/year billed annually; Enterprise custom |
If what you need is a simpler timer-style log rather than a screen record, the roundup of Mac time trackers is a better place to start.
Your uneasiness about AI that sees your screen is a good instinct, and it gets sharper when you break it into parts. Find out what crosses your screen. Map where frames are stored, where they are read, and what a chat sends out. Check who can sit at your keyboard, pick a retention window, and read the license. On Windows, look at Windows Recall. For audio or Linux, look at screenpipe. On a Mac, if you want an MIT-licensed record that can keep its analysis on your own machine, Dayflow is one option. Whichever you choose, you will know exactly what you agreed to.
It depends on three things more than on the idea itself: where the captured frames are stored, where they are analyzed, and who can open them. A tool that stores and analyzes everything on your own device, behind your login, carries different risks from one that uploads frames to a server. Read the vendor's privacy documentation for each of those three steps before you turn recording on.
Local can describe storage, analysis, or both, and vendors do not always mean the same thing. A tool can keep recordings on your disk but send selected screenshots to a cloud model for analysis. Asking an AI assistant about your history can also send that data off-device. Check storage, analysis and chat separately rather than trusting a single "local" label.
Only if you choose a cloud mode. Dayflow stores full screen recordings and its timeline on your Mac. With Local AI through Ollama or LM Studio, screenshot analysis stays on-device. If you configure your own provider, selected screenshots go to that provider under its terms. With Dayflow Pro, the privacy policy says screenshots are deleted after processing, and Activity Data may be retained for sync.
Microsoft documents Recall only as a Windows feature for Copilot+ PCs, and no Microsoft page lists a macOS version. It also needs specific hardware, including a 40 TOPs NPU, 16 GB of RAM, and Windows Hello Enhanced Sign-in Security with a biometric option. If you work on a Mac, you need a different tool for screen memory.
If your main computer is a Copilot+ PC running Windows 11, Windows Recall is the better pick, because Dayflow runs only on macOS 14 or later. Recall stores snapshots locally, always encrypted, and analyzes them on the PC's neural processing unit. If you work on a Mac, Recall is not an option and Dayflow is the one of the two that applies.
Not under its current license. The vendor describes screenpipe as source-available under the Screenpipe Commercial License, which permits personal non-commercial, non-profit, educational and research use and requires a paid license for commercial use. Versions previously released under MIT remain available under MIT. Dayflow, by comparison, is MIT licensed.
screenpipe captures screen and audio, transcribes system and microphone audio with speaker identification, and stores it in a searchable local history. Microsoft states Windows Recall does not record audio. Dayflow sees only the screen. If you want meeting transcripts in the same history as your screen, screenpipe is the one built around that.
That depends on the tool and on who controls the device. Microsoft says Recall is disabled and removed by default on managed commercial devices. Dayflow has no team reports and keeps its record on the Mac it runs on. Whatever the tool, a work computer belongs to your employer, so read your company's device policy before recording anything personal on it.
Dayflow is free and MIT licensed, with an optional Pro plan at $20 a month that handles cloud analysis for you. The free app works with Local AI on your Mac or with your own cloud provider, though the free app does not include paid API usage, so provider charges apply. It runs on macOS 14 or later.
Yes. In Settings, then Export, you can export a date range as Markdown. That gives you a plain-text file you can read, archive, or paste into another tool. Keep in mind that AI summaries can miss or misinterpret work, so check the exported text before you share it with anyone.
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