Ollama Markdown editor on Mac: the setup
Ollama runs a language model straight on your Mac. Connected to Marko’s assistant, it reworks your passages with no remote key and without the text leaving the machine. Here is how to set it up, and what depends on the model and your computer.
Why a local model
With Ollama, the model runs on your Mac. The selection you hand to the assistant goes to no server: it is processed locally, then the proposal comes back in Marko’s difference screen. You reread the old and new text, and you apply or cancel, as with any provider.
This helps when the document holds names, figures or decisions you would rather keep on your machine. In return, your computer does the work: quality and speed depend on the model you pick and on what your Mac can handle.
A local model fits short edits well: rephrase a sentence, summarize a paragraph, translate a selection. On those targeted asks, a model that fits in your Mac’s memory does real work, and you always reread its proposal before you write it.
Set Ollama up
- Install Ollama from ollama.com and launch the app.
- Download a model in the terminal, for example
ollama pull llama3.1, or a lighter model if your Mac has little memory. This download needs a connection; after that, the model answers locally. - Check the model is ready with
ollama list, which shows what you have downloaded. - Leave Ollama running; by default it listens on
http://localhost:11434, an address on your machine. - In Marko, choose the “local model” provider and enter Ollama’s local address.
- Highlight a passage, write your instruction, reread the proposed difference, then apply or cancel.
Pick a model for your Mac
| What you notice | Thing to try |
|---|---|
| Replies take a while | A lighter model, or shorter selections |
| Proposals miss the mark | A larger model if memory allows, or a more precise instruction |
| Memory fills up | Close other apps, or take a lighter model |
| The result varies from one run to the next | Always reread the difference before you apply |
Speed depends on the model and the Mac. There is no latency figure that holds for every machine: speed and quality come down to the model you load, the memory and the chip in your computer. Try it on your own Mac before you decide.
Rereading stays the rule
Lighter models ask for less memory, but they sometimes return less reliable proposals: a misreading, an awkward turn of phrase, an invented detail. That is why the difference always comes before the apply. You read the old and new text, and you apply only what holds up. A local model changes nothing about that rule: whether it runs on your Mac or behind a provider key, the assistant proposes and you decide.
What Marko does not bring
Marko runs on Apple Silicon Macs, from macOS 12 on. Windows, Linux and Intel Mac builds are not distributed yet, and there is no AI service managed by Marko: you bring your own local model or provider key. Ollama and its model install on your side, as described above.
Frequently asked questions
Does my text leave the machine with Ollama?
No. With a local model via Ollama, the selection is processed on your Mac and goes to no server.
Which model should I pick?
One that fits in your Mac’s memory. Lighter models ask for less memory, but they sometimes return less reliable proposals; try it on your own machine before you decide.
Is Ollama fast?
Speed depends on the model and the Mac. There is no latency figure that holds for every machine: speed and quality come down to the model, the memory and the chip.
Does the model need the internet to answer?
No. Downloading the model needs a connection; after that, the model answers locally, by default at http://localhost:11434.
Try it on your own file. Open a .md your AI just wrote, fix it and export it. Marko runs on Apple Silicon Macs, macOS 12 or later.
Full 14-day trial, no account or credit card. See pricing