Guide

How to pick open-source AI apps for your Mac

Updated November 2026 · by Jerry Liu, Dayflow's founder

Three open-source Mac apps are worth knowing. Pick Ollama (MIT) if you want to run open models on your own Mac and chat or build with them. Pick ActivityWatch (MPL-2.0) for a free, local log of apps and window titles with no AI. Pick Dayflow (MIT, Mac only) if you want AI to turn periodic screen captures into a written timeline, with an optional local-AI mode.

You start with a simple wish: some AI on your Mac that doesn’t ship everything you type to someone else’s server. An hour later you have nine tabs open. One app calls itself open source but the repository license is a page of custom terms. Another wants you to pick a model by a name that looks like a serial number. A third installed a menu bar icon you no longer remember adding.

The hard part is rarely the installing. It’s knowing which of these apps you will still be glad to have in a month, what they cost once you leave the free tier, and where your words and screen contents actually go.

This guide covers how to sort that out with a pen, your Mac’s settings and a calendar, and then names three open-source apps worth knowing: Ollama, ActivityWatch and Dayflow. Dayflow is our product, so read that part with that in mind. Everything said about each app comes from the vendors’ own documentation as checked in early October 2026; this is a comparison of published docs, not a lab test.

The short version

Why is it so hard to tell which AI apps are worth keeping?

Partly because “open source” gets used loosely. Some apps publish their code under a license that limits how you can use or redistribute it. That is often called source-available, and it is not the same thing as an open-source license. You can read the code, but your rights are narrower. If openness matters to you, the license file matters more than the marketing page.

Partly because “local” gets used loosely too. An app can run on your Mac and still send things off it, for example when it calls a cloud model for the heavy lifting. Many apps offer both modes, and the default isn’t always the one you’d pick.

And partly because AI apps are new enough that most people haven’t built a sense of what they actually use. It’s common to install something exciting, use it twice, and forget it is running. That isn’t a character flaw. It’s what happens when you choose tools before choosing the job.

The steps below are mostly about getting clear on that job, so the software choice becomes obvious.

Write down the one job you want AI to do

Take a sheet of paper and finish the sentence: “I want AI on my Mac so that I can ___.” Then write one example of when you would have used it in the last three days. If you can’t name a real moment, that’s useful information too.

Be specific about the input. “Summarize long PDFs from work without uploading them” is a different job from “remember what I worked on yesterday”, and those lead to different apps.

Illustration: someone writes “ask questions about client documents without sending them anywhere” and remembers Tuesday, when they pasted a contract into a web chatbot and felt uneasy afterward. Their job is a local model they can chat with. A time tracker would not help them at all.

How to tell after a week: look at the paper again. If the job still feels right and you hit the example situation at least once, you’ve found the real need. If you never hit it, hold off installing anything.

Check your Mac before you download anything

Click the Apple menu, then About This Mac. Note three things on the same sheet of paper: the chip (Apple M series or Intel), the macOS version, and, from System Settings → General → Storage, how much free space you have.

This matters more for AI apps than most software. Ollama’s docs list macOS Sonoma (version 14) or newer. On M series Macs it uses both the CPU and the GPU; on Intel Macs it runs on the CPU only. Its documentation also notes that models are tens to hundreds of GB in size, so disk space can be the first wall you hit. Ollama does not publish a RAM minimum or memory guidance per model size, so if you see a rule of thumb quoted elsewhere, treat it as someone else’s guess and check Ollama’s documentation for anything current.

Dayflow also needs macOS 14 or later.

Illustration: someone on an older Intel MacBook with little free space reads this and decides to try a small model only, or to skip local models for now and keep their AI use to documents they are comfortable sending to a cloud service.

How to tell after a week: if you never had to delete files to make room and your fan didn’t become background noise you resent, your Mac is a fine fit for what you picked. If either happened, scale down.

Read the license and the data line yourself

For any app you are considering, find two things: the license name in its code repository, and one sentence in its docs that says where your data is stored. Write both down.

You are looking for a recognised license name, not a custom one. Of the three apps in this guide, Ollama is MIT licensed, ActivityWatch is MPL-2.0 and Dayflow is MIT. All three let you read the code, run it, and change it.

For data, the vendors say the following. Ollama states that when you run locally, it doesn’t see your prompts or data, and that on macOS models are stored at ~/.ollama/models. ActivityWatch states that all data is stored on your own device and never sent to any server. Dayflow stores full screen recordings and the local timeline on your Mac, and where its AI analysis happens depends on the mode you choose (more on that below).

Illustration: someone finds an AI app they like, opens the repository, and sees a license with the word “commercial” in a restriction clause. They keep using it but stop calling it open source in their notes, and they look for an MIT or MPL alternative for the work they care about most.

How to tell after a week: you should be able to explain, out loud and without looking, where each app you kept puts your data. If you can’t, go back to the docs for that one.

Run models on your own Mac with Ollama

If your job from the first step involves chatting with a model, drafting text, or asking questions of documents, a local model runner is the core piece. Ollama runs open models on your machine through a command line tool, a desktop app and a local API, with Python and JavaScript libraries if you want to build on it. On Apple devices it uses Metal for GPU acceleration.

Local use is free. Ollama also offers optional cloud-hosted models that need no download and no GPU. The Free plan includes starter cloud credits and starter cloud models; heavier cloud use needs a paid plan. Pro is $20 a month or $200 a year billed annually, Max is $100 a month, Team is $500 a month in early access, and Enterprise is custom. The pricing page lists no trial.

For cloud models, Ollama says it processes prompts and responses to provide the service but does not store or log that content. If you want to be certain nothing leaves your Mac, the docs say you can turn cloud features off with OLLAMA_NO_CLOUD=1.

Illustration: someone installs the desktop app, downloads one model, and keeps a sticky note listing the three things they asked it that day. They set the environment variable to disable cloud features because their job was “nothing leaves the machine.”

How to tell after a week: count the sticky notes. If you used it most days for the job you wrote down, it has earned its disk space. If the notes are blank after day two, delete the model files and reclaim the space.

Keep a plain record before you add AI to it

If your job is closer to “understand where my time goes,” start without software. For two working days, keep a paper log next to your keyboard. Each time you switch to something different, write the time and a few words. A daily work log template can give this some shape if a blank page feels hard.

Many people find that two days of this shows the pattern they suspected, and sometimes one they didn’t. Many also find that writing every switch is tiring, which is the honest reason automatic tools exist.

If you want the automatic version with no AI involved, ActivityWatch is the open-source option to know. It records the active app, the window title and whether you are away from the keyboard. Browser extensions add the tab title and URL, and editor plugins add coding activity. It is free with no feature gates, keeps everything on your device, and runs on Windows, Linux and Android as well as Mac. It lists no AI summaries, sync between devices is still a work in progress, and there is no iOS version. We compare it side by side with our own app in ActivityWatch vs Dayflow.

If you’d rather survey the wider field of Mac time trackers, including paid ones, that roundup goes through them.

Illustration: someone keeps a paper log Monday and Tuesday, notices that most of their afternoon disappears into one chat app, and installs an automatic tracker on Wednesday to confirm it without having to write things down.

How to tell after a week: you should be able to answer “where did Thursday afternoon go?” from your record, not from memory. If you still have to guess, the record isn’t capturing what you need.

Decide where your line is for screen data

Some AI apps look at what is on your screen. That is powerful and it deserves a deliberate choice. On paper, list what regularly appears on your screen during a working week: banking, health portals, private messages, client files, password managers, other people’s personal information.

Next to each, write “fine on my Mac”, “fine on my Mac only”, or “should never be captured.” That list tells you which processing modes you can accept and which apps or windows you may want to keep off-screen while a capture tool is running.

Illustration: a freelancer writes that client files are “fine on my Mac only,” so any screen-reading AI they use has to run locally, never through a cloud model.

How to tell after a week: if you felt a flicker of worry each time a certain window opened, that window belongs in a stricter column.

Give each app one week and a stop rule

Put a calendar event seven days after any install, titled with the app’s name and the question “keep, change, or remove?” In the notes, paste the job you wrote in the first step.

When the reminder fires, answer with one of three words. Keep means it did the job. Change means it did part of the job and one setting would fix it. Remove means you’ll uninstall it now, including any large model files.

Illustration: someone installs three AI apps in one weekend, sets three reminders, and keeps one. The other two would otherwise have kept running in the background for months.

How to tell after a week: this one is self-checking. If the reminder fires and you make the call in under a minute, it’s working.

Where seeing your actual day helps

Some jobs from the first step are about memory: what did I actually do today, this week, on that project? A paper log works until the day you forget to write in it, and reconstructing from memory tends to be generous to the parts you remember and quiet about the rest. If that sounds familiar, where did my day go covers the problem in more depth.

Dayflow is an open-source Mac app (MIT, macOS 14 or later) that records the screen at one frame every 10 seconds and has AI write the day into a readable timeline. You don’t start a timer. Full recordings and the local timeline are stored on your Mac.

The AI analysis runs in one of three modes. Local AI runs a supported model on your Mac through Ollama or LM Studio, so screenshot analysis stays on-device, using your own hardware. Your own provider sends selected screenshots to the cloud provider you configure, under its terms and charges. Dayflow Pro, $20 a month, manages cloud analysis; under the privacy policy it deletes screenshots after processing and may retain Activity Data for sync. The free app is $0. You can export a date range as Markdown from Settings → Export. More detail is on local and cloud processing.

Dayflow publishes this article, so weigh this section with that in mind.

Where Dayflow is the wrong pick

If you want to think through the privacy side more generally, time tracking without surveillance and is Dayflow safe go further.

The bottom line

Pick the job first, then the app. If the job is talking to a model without sending your words anywhere, Ollama is the open-source place to start, and local use costs nothing. If the job is a plain, private record of which apps you used, ActivityWatch does that for free on almost any computer. If the job is a written account of your Mac day without remembering to log it, that’s what Dayflow is for, with a local-AI option if you want analysis to stay on your machine.

Whatever you choose, check the license, know where the data lives, and set a one-week reminder. The apps worth keeping will still be useful when it fires.

Frequently asked questions

Is Ollama free to use on a Mac?

Running models locally with Ollama is free. The Free plan also includes starter cloud usage credits and starter cloud models. Paid plans are for heavier cloud use: Pro is $20 a month or $200 a year billed annually, Max is $100 a month, Team is $500 a month in early access, and Enterprise is custom. Ollama's pricing page lists no trial.

Does Ollama send my prompts to the cloud?

Not when you run models locally. Ollama states that it does not see your prompts or data when you run locally. Cloud-hosted models are different: the vendor says it processes prompts and responses to provide the service but does not store or log that content. You can switch cloud features off entirely by setting OLLAMA_NO_CLOUD=1.

What Mac do I need to run Ollama?

Ollama's macOS documentation lists macOS Sonoma (version 14) or newer. On Apple M series Macs it uses both the CPU and the GPU, through Metal. Intel Macs are supported but run on the CPU only. The docs state no RAM minimum, so check the vendor's documentation and model pages for current requirements before downloading a large model.

How much memory do I need to run local AI models on a Mac?

Ollama does not currently publish memory-per-model-size guidance in its README or its macOS, GPU or FAQ docs, so any rule of thumb you see quoted is not from the vendor. Check Ollama's documentation at the time you install, and keep an eye on Activity Monitor during your first week to see how your own Mac handles the models you actually use.

Is ActivityWatch open source and free?

Yes. ActivityWatch is licensed under MPL-2.0 and the vendor describes it as completely free, with no feature gates, usage limits or ads. It records your active app, window title and away-from-keyboard time, and browser extensions add tab titles and URLs. The vendor states all data stays on your device. It lists no built-in AI summaries.

Is Dayflow open source?

Yes. Dayflow is MIT licensed and free, with an optional Pro plan at $20 a month that manages cloud analysis for you. It runs on macOS 14 or later, records the screen at one frame every 10 seconds, and has AI write the day into a readable timeline. Recordings and the timeline are stored on your Mac.

Can Dayflow use local AI instead of the cloud?

Yes. In Dayflow's Local AI mode, a supported model runs on your Mac through Ollama or LM Studio, and screenshot analysis stays on-device. Setup and local inference use your own hardware. The other options are your own cloud provider, which receives selected screenshots under its own terms, or Dayflow Pro, which deletes screenshots after processing under its privacy policy.

What is a better tracker than Dayflow if I use Windows or Linux?

ActivityWatch. Dayflow is Mac only, while ActivityWatch runs on Windows, Mac, Linux and Android, is free under MPL-2.0, and keeps its data on your own device. It logs apps and window titles rather than screen captures, and it lists no AI summaries, but if you need one tool across several operating systems it is the clear pick.

Does ActivityWatch have AI summaries?

No AI features are listed by ActivityWatch. It gives you a local record of which apps and windows were active, plus browser and editor data if you add the extensions and plugins. Some people pair a plain activity log like this with a daily written note. If you want AI-written summaries of your day, that is a different kind of tool.

See where your day actually went

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