Guide
Install Ollama or LM Studio, download an open model, and run it on your own Mac. Pick Ollama if you want an MIT-licensed tool with a command line, desktop app and local API; it runs on macOS 14+, and Intel Macs are CPU only. Pick LM Studio for a free desktop app with MLX support; it needs Apple silicon and macOS 14+, with 16GB+ RAM recommended.
You paste a paragraph from a client contract into an AI chat window, then stop before pressing Enter. Or you open a draft of something personal, think “I’d like help with this”, and then wonder where that text ends up after you send it. Sometimes it is simpler than that: you are on a train with no signal and the chat tab just spins.
That is usually when the question comes up: can the model run on my own Mac instead? It can. Two free apps, Ollama and LM Studio, download an open model to your disk and run it on your Mac’s own chip. This guide covers what “local” actually means, how to check your Mac before downloading anything, how to choose between the two, and how to tell after a week whether the setup is worth keeping.
A note on who is writing: Dayflow is our product, and it can use either of these apps for its own on-device analysis. Everything below about Ollama and LM Studio comes from their own documentation as checked on October 4, 2026. Nothing here was benchmarked, so you will not find speed figures or model recommendations that the vendors do not publish.
OLLAMA_NO_CLOUD=1.A chat model is, at bottom, a very large file. A cloud assistant keeps that file on someone else’s servers: you send text over the internet and get text back. Running locally means the file sits on your own disk and a program on your Mac turns your prompt into an answer.
So there are two pieces. The runner is the app, Ollama or LM Studio. The model is the open model you download through it. The runner is the small part. The models are not: Ollama’s documentation describes them as tens to hundreds of GB in size, which is why free disk space matters more than most people expect.
Local AI feels different from cloud chat in a few ways that many people notice quickly:
One more wrinkle: both vendors now also offer cloud-hosted models. Saying “I use Ollama” or “I use LM Studio” no longer automatically means every prompt stays on your Mac. The last step below deals with that.
What to do: before installing anything, take a sheet of paper and list three to five specific jobs. “Help me write” is too vague to test. “Turn my rough meeting notes into five bullets” is a job. Next to each one, mark whether the text involved is private enough that you would rather it never leave your Mac.
Illustration: a list might read: “1. Tidy meeting notes (private, client names). 2. Reword awkward emails (sometimes private). 3. Explain an error message (not private). 4. Summarize a long PDF I was sent (private).” Two of those clearly benefit from staying local. One does not need it at all.
How to tell after a week: look at the list and circle the jobs you actually used the local model for. If you circled none, the problem is probably not the model or the app. It may be that cloud chat already covers what you do, or that the jobs were too vague to reach for.
What to do: open the Apple menu and choose About This Mac. Write down the chip, the macOS version and the memory. Then open System Settings → General → Storage and note your free disk space. No extra software needed.
Compare those notes with what each vendor states:
Ollama’s macOS, GPU and FAQ docs state no RAM minimum. You may come across rules of thumb online that pair amounts of memory with model sizes, but they are not in Ollama’s current documentation, so do not treat them as official. Read the page for the specific model you want instead.
Illustration: your note says “Apple silicon, macOS 15, 8GB memory, 60GB free.” That rules neither app out. It does tell you LM Studio’s own guidance points you toward smaller models, and that a couple of large downloads could eat most of that free space.
How to tell after a week: you have not seen a “disk almost full” warning, and you have not downloaded a model your Mac could not load. If either happened, go back to the note and check what you skipped.
What to do: install one app, not both. Switching between them in the first few days mostly produces two sets of downloaded models and no real opinion about either.
| Ollama | LM Studio | |
|---|---|---|
| Mac requirements | macOS 14+; Apple M series gets CPU and GPU, Intel is CPU only | macOS 14.0+, Apple silicon only |
| Mac memory guidance | No RAM minimum stated in the docs | 16GB+ recommended; 8GB may work with smaller models |
| Other platforms | Windows, Linux, Docker | Windows (x64 and ARM), Linux (x64 and ARM64) |
| How you use it | Command line, desktop app, local API, Python and JavaScript libraries | Desktop app with the Bionic agent; headless daemon (llmster) for servers |
| License | Open source (MIT) | Proprietary |
| Running models locally | Free | Free, for home and work |
| Paid plans | Pro $20 a month or $200 a year billed annually; Max $100 a month; Team $500 a month (early access); Enterprise custom | Bionic+ $20 USD a month; Pro $100 USD a month |
| What paid plans pay for | Cloud usage credits | Cloud inference and web search |
| Free trial | None stated | None stated |
A few details that help decide. Ollama’s vendor describes it as a way to use open models in desktop apps and coding agents, or to build them into an application, and its local API is the reason many people pick it. Its paid plans are metered in cloud usage credits: the pricing page lists $60 of credits a month on Pro, $300 on Max and $1,000 shared on Team.
LM Studio runs models with llama.cpp on every platform and adds Apple’s MLX on Apple silicon Macs. Its free plan includes the Bionic agent, offline voice transcription, LM Link for up to five devices and limited web search. The vendor’s pricing page is now branded “LM Studio Bionic”, while the download page still says LM Studio. It is not open source, and its terms prohibit modifying or redistributing the app.
Illustration: someone who mostly wants a chat window and never opens Terminal installs LM Studio. Someone who wants a coding tool or script to call a model on their own machine installs Ollama for its local API.
How to tell after a week: you opened the app on more days than not, and you can name one thing it did that you would miss. If you spent more time reading comparisons than using it, only switch if a specific missing feature actually blocked you.
What to do: browse the models your runner offers, start with a smaller one, and read its page for size and requirements before downloading. Download one model, not five. Forum threads are full of confident advice that was written for different hardware, a different version or a different job.
Illustration: you find a model whose page lists a size that fits comfortably in your free disk space. You download it, run one prompt from your list, and leave the rest of the catalog alone until the test in the next step is done.
How to tell after a week: the model loads without your Mac struggling, which in practice means loud fans, other apps slowing down or the machine pausing while it works. If it struggles, step down to a smaller model before blaming the app.
What to do: on an index card or a sheet of paper, copy five prompts from the list you wrote in the first step. Use real text, not “write me a poem”. Run all five on each model you try and score each answer: 1 for not usable, 2 for usable with edits, 3 for usable as-is. If you want a number for how long you waited, use a phone stopwatch or a kitchen timer, and note only whether the wait felt fine for that job.
Illustration: your card reads “meeting notes: 3, email rewrite: 3, error message: 2, PDF summary: 1, client recap: 2.” That is a clear picture. The model handles short rewrites well and struggles with the long document.
How to tell after a week: the card tells you what to do next. Keep the model for the jobs that scored well, try a different model for the ones that did not, or accept that a certain job is better done somewhere else. A mixed result is normal and useful.
What to do: write a one-line rule for yourself, then set up the app to match it. This matters because both apps offer cloud models alongside local ones.
OLLAMA_NO_CLOUD=1.Illustration: “Client documents: local model only. Everything else: whatever works best.” Written down, a rule like that is easy to follow. Kept in your head, it tends to slip on a busy afternoon.
How to tell after a week: for any prompt you sent, you can say where it was processed. A simple check that needs no software: turn off Wi-Fi and run your usual prompts. If everything still works, your workflow is local.
Once a model runs on your Mac, one quietly useful job for it is answering where did my day go without you having to remember to start a timer or rebuild the day from memory at 6pm.
Dayflow records your screen at one frame every 10 seconds and has AI write the day into a readable timeline. Full screen recordings and the timeline are stored on your Mac. In its Local AI mode, a supported model runs on your Mac through Ollama or LM Studio, so screenshot analysis stays on-device. Setup and local inference use your own hardware resources.
There are two other modes. You can send selected screenshots to your own cloud provider, whose charges and privacy terms apply. Or you can use Dayflow Pro at $20 a month, which manages cloud analysis; under the privacy policy, Pro deletes screenshots after processing and may retain Activity Data for sync. The differences are laid out in local and cloud processing. One caveat: if you ask a cloud assistant about your Dayflow data, that conversation is not local just because the recordings are.
You can export a date range as Markdown from Settings → Export. The app is free, MIT licensed and runs on macOS 14 or later.
Dayflow publishes this article, and Dayflow is the product described in this section.
Running AI locally on a Mac comes down to a runner and a model. Check your chip, macOS version, memory and free disk space first. Then pick one runner: Ollama if you want MIT licensing, a local API and support for Intel Macs, LM Studio if you want a free desktop app with MLX on Apple silicon. Start with one smaller model, judge it against five of your own prompts, and decide in writing which prompts are allowed to leave the machine. After a week you will know whether local AI earns its disk space, and for which jobs.
Yes. Ollama is free for local use and MIT licensed, and LM Studio is free for both home and work use. Both need macOS 14 or newer. Their paid plans pay for cloud-hosted models, usage credits or web search, not for running models on your own Mac. Your real costs are disk space, memory and the time it takes to set things up.
It depends on how you work. Ollama is MIT licensed and offers a command line, a desktop app, a local API and Python and JavaScript libraries, and it also runs on Intel Macs, CPU only. LM Studio is a proprietary but free desktop app that supports both llama.cpp and Apple's MLX, and it requires Apple silicon and macOS 14.0 or newer.
No. LM Studio's system requirements state that Macs need Apple silicon and macOS 14.0 or newer, and Intel Macs are not supported. If you have an Intel Mac on macOS Sonoma or newer, Ollama is the option of the two: its docs say x86 Macs run models on the CPU only, while Apple M series Macs get both CPU and GPU support.
LM Studio recommends 16GB or more of RAM on a Mac and says 8GB machines may work with smaller models. Ollama's macOS, GPU and FAQ docs state no RAM minimum, and the memory rules of thumb you may see online are not in its current documentation. Check the page for the specific model you plan to download before you pull it.
On macOS, Ollama stores downloaded models at ~/.ollama/models in your home folder. That matters because the vendor describes models as tens to hundreds of GB in size, so a few downloads can take a large share of your disk. Check free space in System Settings before pulling a new model, and delete the ones you no longer use.
Not when the model runs locally. Ollama states that it does not see your prompts or data when you run locally, and LM Studio says it can operate entirely offline once model files are downloaded. Both also offer cloud-hosted models, though, and prompts sent to those leave your Mac. Ollama's cloud features can be switched off with OLLAMA_NO_CLOUD=1.
Ollama is open source under the MIT license, as published in its main GitHub repository. LM Studio is not open source: it is a proprietary desktop app, and its terms prohibit modifying, redistributing or reverse engineering the software. LM Studio is still free to use at home and at work, with no separate commercial license needed.
Ollama lists Pro at $20 a month or $200 a year billed annually, Max at $100 a month, Team at $500 a month in early access, and custom Enterprise pricing. LM Studio lists Bionic+ at $20 USD a month and Pro at $100 USD a month. In both cases the paid tiers cover cloud usage, not running models locally.
Yes. Dayflow's Local AI mode runs a supported model on your Mac through Ollama or LM Studio, so screenshot analysis stays on-device. Setup and local inference use your own hardware. You can instead send selected screenshots to your own cloud provider or use Dayflow Pro at $20 a month. Dayflow itself is free, MIT licensed and needs macOS 14 or later.
Use LM Studio, or Ollama, for that. Both exist to run open models you can talk to, and Ollama adds a local API for coding agents and apps. Dayflow is a different kind of tool: it records your screen and has AI write your day into a timeline. It can use either runner for that analysis, but it is not a general chat app.
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