Guide
With bring-your-own-key, the AI provider bills you per use and its terms govern what you send. In Dayflow, Local AI through Ollama or LM Studio wins when screenshots must stay on your Mac. Your own provider fits if you already accept its terms and charges. Dayflow Pro, at $20 a month, wins when you want a set monthly price instead of usage billing.
You install an app that promises to summarize your day, tidy your notes or explain what’s on your screen. Partway through setup there’s a text field: “Paste your API key.” You open an AI provider’s website, make an account, add a card, copy a long string of characters, paste it in, and the app starts working. It feels like a one-time chore, like entering a license code.
A few weeks later a receipt arrives from a company you barely remember signing up with. The amount might be tiny. It might be more than the app itself would have cost. Then a quieter question shows up: that app was looking at your screen, your documents, your messages. Where did all of it go, and who kept a copy?
This guide explains how bring-your-own-key (BYOK) apps work, the two bills and two sets of terms you’ve actually signed up for, and six things you can do this week, mostly with paper and a calendar, to know what you’re paying and who sees what. Dayflow, the Mac app that publishes this guide, offers a local option, an own-key option and a paid plan, so it serves as the worked example near the end. Dayflow details come from version 2.2.0 and its published policy as reviewed on September 6, 2026.
An API key is a credential. It lets a piece of software make requests to an AI provider’s models and charge them to your account. The app sends text or images, the provider sends back an answer, and the provider records the usage against your key. The app maker may charge you nothing, a one-time fee or a subscription, but the AI work itself is billed by the provider.
That’s why so many apps, including a growing number of Mac time trackers, note-taking tools and writing apps, now offer a key field. It lets the app maker skip the cost of running AI for everyone, and it lets you pick a provider you already use.
Providers generally bill by how much goes in and comes out. A short sentence is small. A long document is larger. Screenshots and other images are usually larger still. You rarely see these units while using the app, and the provider’s price list uses terms that don’t map neatly onto “summarize my morning.”
Apps that run in the background make it harder. A tool that analyzes your work through the day keeps sending requests while you’re busy doing something else. Retries after errors, longer prompts as the app adds context, and higher-quality model settings all add to the count without any visible signal. Nothing shows the app’s price and the provider’s usage side by side, so the total only becomes clear when the bill arrives.
The path usually looks like this: your Mac, then possibly the app maker’s servers, then the provider. Some apps send requests straight from your computer to the provider. Others relay them through their own servers first. Those are very different arrangements, and the setup screen rarely says which one you’re in.
Once data reaches the provider, the provider’s terms govern it: how long it’s logged, whether it can be used to improve models, which staff can look at it and in what situations, and which country it sits in. The app maker’s privacy policy can tell you what the app does, but it can’t make promises on the provider’s behalf.
Setup happens once, often in a hurry. Cost and data flow happen continuously and out of sight. The key field looks like a preference, but it’s closer to a contract, and the terms you agreed to were usually on a signup page you clicked through to get the key. Many people only start asking questions after the first surprise. That’s normal, and it’s fixable.
What to do: Take one sheet of paper and draw three boxes: your Mac, the app maker, the AI provider. Draw an arrow for each kind of data the app might send: screenshots, typed text, file contents, finished summaries. Label each arrow “always,” “sometimes” or “never.” Fill it in from the app’s settings and privacy page. Where you can’t tell, write a question mark and email the app’s support with that specific question.
Illustration: Say you use a meeting-notes app with your own key. Your drawing might show an “always” arrow from your Mac to the provider for transcripts, a question mark on whether transcripts pass through the app maker’s server, and a “never” arrow for the audio itself because the settings say audio is transcribed on-device. That one question mark becomes your support email.
How to tell after a week: The question marks are either answered or turned into decisions (“I won’t use this for client calls until I hear back”). A good test is whether you can explain the drawing to a friend in about a minute.
What to do: Write four questions on an index card and answer each one from the provider’s own policy pages:
Also check which terms apply. Providers often have separate terms for developer or API accounts and for their consumer chat products, and an API key usually falls under the developer terms. Next to each answer, copy the sentence it came from and the date you read it.
Illustration: Your card might read: “Kept for a limited period for abuse monitoring (quote, date). Not used for training on API traffic by default (quote, date). Staff review only for flagged content (quote, date). Region: couldn’t find.” That last line is useful information in itself.
How to tell after a week: Every question has an answer or an explicit “unknown.” Treat any unknown as the less comfortable possibility and decide what you’ll keep away from that app until you find out.
What to do: Before the first bill, write down a rough estimate:
requests per hour × hours the app runs per day × working days per month × price per request
The price per request comes from the provider’s pricing page combined with what the app’s documentation says it sends. For background apps, count every hour the app is running, not just the hours you’re actively using it. If the app sends images, use the image pricing, not the text pricing.
Illustration: Picture a screen-summary app that sends a batch of screenshots every few minutes across an eight-hour workday. Work out the batches per day, multiply by the provider’s listed price for that kind of request, then by your working days. The answer doesn’t need to be exact. It needs to be in the right range so the first bill isn’t a shock.
How to tell after a week: Compare your estimate with the provider’s usage page, scaled to a week. If they’re roughly in line, you can trust your estimate going forward. If they’re far apart, the app is sending more, or bigger, requests than you assumed, and that’s worth asking the app maker about.
What to do: If the provider account offers a spending limit or a usage alert, set one at a number you’d be calm about paying. Then add a recurring calendar event for Friday afternoon: five minutes, “check AI usage.” Open the provider’s usage page and write the week’s number in a notebook. After the first month, move the reminder to monthly.
Illustration: A notebook page with a column of four Friday numbers and a note beside the third one: “jumped, turned on the higher-quality setting Tuesday.” That note is the whole point. It connects a change in the bill to something you did.
How to tell after a week: You have one number written down and you know exactly where to find the next one. If the calendar reminder got dismissed without the check, move it to a time you’re actually at your desk.
What to do: Create a separate key for each app and name it after the app in the provider’s dashboard. Never paste a key into an email, a chat, a shared document or a screenshot. Keep a short list, in a password manager or on paper at home, of which apps hold which keys. When you stop using an app, revoke its key the same day.
Illustration: Your list reads “notes-app key, created March; writing-app key, created June; old summary app, revoked September.” If one app starts behaving oddly or its usage spikes, you shut off that single key and everything else keeps working.
How to tell after a week: You could cut off any one app’s access in a minute without touching the others, and you know which key belongs to which app without guessing.
What to do: Sit down for fifteen minutes, alone or with a colleague who handles similar work, and list the kinds of material you’d be uncomfortable sending to an outside company: client documents under confidentiality agreements, health information, other people’s private messages, passwords that show up on screen. For each item, decide what the app is allowed to do with it. If the app sends to a cloud provider, your options are to pause it during that work, switch to a mode that runs the AI on your Mac, or not use the app for that work at all.
Remember the second-hand routes too. If you export something from an app that stores data locally and paste it into a cloud chat tool, the export now goes to that chat tool’s provider. Asking a cloud assistant about your local data works the same way.
Illustration: Your list says “client contracts: pause the app while reviewing; health portal: never on screen while the app runs; team chat: allowed.” A sticky note on the monitor with “pause first” is enough of a system for many people.
How to tell after a week: There was no moment where you thought “oh, that was on screen” with no plan for it. If there was, add it to the list.
Dayflow is a free, open-source Mac app (MIT license, macOS 14 or later) that records your screen at one frame every 10 seconds and has AI write the day into a readable timeline. You don’t have to remember to start a timer or reconstruct where your day went from memory. Full screen recordings and the timeline are stored on your Mac. You choose who does the AI analysis:
| Mode | Where analysis happens | What you pay | Terms that apply |
|---|---|---|---|
| Local AI | On your Mac, through Ollama or LM Studio | $0, but it uses your own hardware | Screenshot analysis stays on-device |
| Your own provider | Selected screenshots go to the provider you configure | $0 for the app, plus your provider’s charges | The provider’s own terms |
| Dayflow Pro | Dayflow manages cloud analysis | $20 a month | Dayflow’s privacy policy: screenshots deleted after processing, Activity Data may be retained for sync |
The free app does not include paid API usage. An assistant can receive the Dayflow data it retrieves in your conversation, so local storage does not make a cloud assistant conversation local. You can export a date range as Markdown from Settings → Export. The local and cloud processing page and pricing have the details.
Dayflow publishes this article, so weigh what’s said here with that in mind.
When you paste in a key, you take on a second company: one that bills you by usage and handles your data under its own terms. Neither fact is a problem on its own. The problem is not knowing them.
Draw what leaves your machine. Answer the four questions about the provider’s terms. Estimate the bill before it arrives, cap it if you can, and check it weekly for a month. Give every app its own key. Decide in advance what stays on your Mac.
If you’d rather not run a second account at all, pick an app with a local mode or a flat plan, and read its policy with the same index card.
It means the app does not include the AI work in its own price. You create an account with an AI provider, generate a key, and paste it into the app. The app then sends requests to that provider on your account. The provider bills you for that usage under its own pricing, and its own terms decide how it handles what the app sends.
It depends on how much the app sends. Usage billing can cost very little for light use and more than a flat plan for an app running all day. Estimate first: multiply requests per day by the provider's listed price and your working days. In Dayflow, the app is $0 and your provider charges are separate, while Dayflow Pro is a flat $20 a month.
At minimum, the AI provider receives whatever the app sends with your key, and its terms decide retention, training use and staff access. The app maker may also see data if it relays requests through its own servers. Check both privacy policies, and ask the app maker directly whether requests go straight from your device to the provider.
No. Your own key changes who bills you and whose terms apply, not whether data leaves your machine. The content still goes to the provider's servers. If you need analysis to stay on your computer, use an app mode that runs the AI model locally. In Dayflow, that is the Local AI mode, which keeps screenshot analysis on-device.
That depends entirely on the provider's terms, and terms for developer or API accounts can differ from terms for consumer chat products. Look for a plain statement about training use and whether you can opt out. Note the date you read it, because terms change. If you cannot find a clear answer, treat the data as if it could be used.
Yes. Dayflow offers Local AI, where a supported model runs on your Mac through Ollama or LM Studio and screenshot analysis stays on-device. That needs your own hardware resources and some setup. The other no-key option is Dayflow Pro at $20 a month, where Dayflow manages cloud analysis. Your own provider is a third option for people who already have a key.
Selected screenshots are processed off-device by Dayflow. Under Dayflow's privacy policy, Pro deletes screenshots after processing and may retain Activity Data for sync. This is a policy commitment rather than something you can verify from your Mac. Full screen recordings and the local timeline stay stored on your Mac in every mode. The privacy page has the full wording.
No. An assistant can receive the Dayflow data it retrieves in your conversation, and a cloud assistant handles that data under its own provider's terms. Local storage of the timeline does not make the conversation local. The same applies if you export a date range as Markdown and paste it into a cloud chat tool.
No, and here another tool beats it. Dayflow does not provide website blocking, invoices, payroll approvals or team reports. If you need sites blocked during work hours, a dedicated website blocker is the better pick. If you bill clients, invoicing or billing software is. Dayflow is a record of your screen time, not a blocker or a billing system.
Revoke the key in your provider account right away, then create a new one and paste it only into the app that needs it. Check the provider's usage page for requests you do not recognize and contact the provider about charges you did not make. Using one key per app makes this faster, because revoking one key does not break your other apps.
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